Artificial Intelligence

Using Federated Learning

Federated machine learning is a novel approach to combining distributed machine
learning, cryptography, security, and incentive mechanism design. It allows
organizations to keep sensitive and private data on users or customers decentralized
and secure, helping them comply with stringent data protection regulations like
GDPR and CCPA.
Artificial Intelligence Using Federated Learning: Fundamentals, Challenges, and
Applications enables training AI models on a large number of decentralized devices
or servers, making it a scalable and efficient solution. It also allows organizations to
create more versatile AI models by training them on data from diverse sources or
domains. This approach can unlock innovative use cases in fields like healthcare,
finance, and IoT, where data privacy is paramount.
The book is designed for researchers working in intelligent federated learning
and its related applications, as well as technology development, and is also of interest
to academicians, data scientists, industrial professionals, researchers, and students.
Intelligent Manufacturing and Industrial Engineering
Series Editor: Ahmed A. Elngar, Beni-Suef University, Egypt.
Mohamed Elhoseny, Mansoura University, Egypt

Machine Learning Adoption in Blockchain-Based Intelligent Manufacturing

Edited by Om Prakash Jena, Sabyasachi Pramanik, Ahmed A. Elngar

Integration of AI-Based Manufacturing and Industrial Engineering Systems

with the Internet of Things
Edited by Pankaj Bhambri, Sita Rani, Valentina E. Balas and Ahmed A. Elngar

AI-Driven Digital Twin and Industry 4.0

A Conceptual Framework with Applications
Edited by Sita Rani, Pankaj Bhambri, Sachin Kumar, Piyush Kumar Pareek, and
Ahmed A. Elngar

Technology Innovation Pillars for Industry 4.0

Challenges, Improvements, and Case Studies
Edited by Ahmed A. Elngar, N. Thillaiarasu, T. Saravanan, and Valentina Emilia
Balas

Internet of Things and Big Data Analytics-Based Manufacturing

Edited by Arun Kumar Rana, Sudeshna Chakraborty, Pallavi Goel, Sumit Kumar
Rana, and Ahmed A. Elngar

Industrial Internet of Things Security

Protecting AI-Enabled Engineering Systems in Cloud and Edge Environments
Edited by Sunil Kumar Chawla, Neha Sharma, Ahmed A. Elngar, Prasenjit
Chatterjee, and P. Naga Srinivasu

Intelligent Manufacturing and Industry 4.0

Impact, Trends and Opportunities
Edited by Alka Chaudhary, Vandana Sharma, and Ahmed Alkhayyat

Artificial Intelligence Using Federated Learning

Fundamentals, Challenges, and Applications
Edited by Ahmed A. Elngar, Diego Oliva, and Valentina E. Balas

For more information about this series, please visit: www.routledge.com/Mathematical-

Engineering-Manufacturing-and-Management-Sciences/book-series/CRCIMIE
 Artificial Intelligence
Using Federated Learning
Fundamentals, Challenges,
and Applications

Edited by Ahmed A. Elngar, Diego Oliva

and Valentina E. Balas

Boca Raton London New York

CRC Press is an imprint of the

Taylor & Francis Group, an informa business
Designed cover image: Shutterstock—luchschenF
First edition published 2025
by CRC Press
2385 NW Executive Center Drive, Suite 320, Boca Raton FL 33431
and by CRC Press
4 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN
CRC Press is an imprint of Taylor & Francis Group, LLC
© 2025 selection and editorial matter, Ahmed A. Elngar, Diego Oliva and Valentina E. Balas;
individual chapters, the contributors
Reasonable efforts have been made to publish reliable data and information, but the author and
publisher cannot assume responsibility for the validity of all materials or the consequences of
their use. The authors and publishers have attempted to trace the copyright holders of all material
reproduced in this publication and apologize to copyright holders if permission to publish in this
form has not been obtained. If any copyright material has not been acknowledged please write and
let us know so we may rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced,
transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or
hereafter invented, including photocopying, microfilming, and recording, or in any information
storage or retrieval system, without written permission from the publishers.
For permission to photocopy or use material electronically from this work, access www.
copyright.com or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive,
Danvers, MA 01923, 978-750-8400. For works that are not available on CCC please contact
mpkbookspermissions@tandf.co.uk
Trademark notice: Product or corporate names may be trademarks or registered trademarks and are
used only for identification and explanation without intent to infringe.
ISBN: 978-1-032-77164-9 (hbk)
ISBN: 978-1-032-77246-2 (pbk)
ISBN: 978-1-003-48200-0 (ebk)
DOI: 10.1201/9781003482000
Typeset in Times
by Apex CoVantage, LLC
Contents
Preface���������������������������������������������������������������������������������������������������������������������vii
About the Editors������������������������������������������������������������������������������������������������������ix
List of Contributors���������������������������������������������������������������������������������������������������xi

Chapter 1 Federated Learning: Overview, Challenges, and Ethical

Considerations�������������������������������������������������������������������������������������� 1
Jagjit Singh Dhatterwal, Kiran Malik, Kuldeep Singh Kaswan,
and Ahmed A. Elngar

Chapter 2 In-Depth Analysis of Artificial Intelligence Practices:

Robot Tutors and Federated Learning Approach in English
Education�������������������������������������������������������������������������������������������� 16
Muthmainnah, Akkarapon Nuemaihom, Ahmad Al Yakin,
Prodhan Mahbub Ibna Seraj, Muhammad Mujtaba Asad, and
Ahmed A. Elngar

Chapter 3 Enabling Federated Learning in the Classroom:

Sociotechnical Ecosystem on Artificial Intelligence
Integration in Educational Practices�������������������������������������������������� 37
Ahmad Al Yakin, Arkas Viddy, Idi Warsah,
Ali Said Al Matari, Luís Cardoso, Ahmed A. Elngar,
Ahmad J. Obaid, and Muthmainnah

Chapter 4 Real-Time Implementation of Improved Automatic Number

Plate Recognition Using Federated Learning������������������������������������ 57
M. Venkatanarayana, Syed Zahiruddin, and Ahmed A. Elngar

Chapter 5 Fake Currency Identification Using Artificial Intelligence

and Federated Learning���������������������������������������������������������������������66
Syed Zahiruddin, Vamsi Krishna Kadiri,
Valli Bhasha Achukatla, Pavan Kumar Kattela, and
Ahmed A. Elngar

Chapter 6 Blockchain-Enhanced Federated Learning for

Privacy-Preserving Collaboration������������������������������������������������������ 83
Pawan Whig, Balaram Yadav Kasula, Nikhitha Yathiraju,
Anupriya Jain, Seema Sharma, and Ahmed A. Elngar
v
viContents

Chapter 7 Federated Learning-Based Smart Transportation Solutions:

Deploying Lightweight Models on Edge Devices in the
Internet of Vehicles�������������������������������������������������������������������������� 112
Sivabalan Settu, Raveendra Reddy, Appalaraju Muralidhar,
Thangavel Murugan, and Rathipriya Ramalingam

Chapter 8 Application of Artificial Intelligence and Federated Learning

in Petroleum Processing������������������������������������������������������������������� 134
Abdelaziz El-hoshoudy

Chapter 9 Artificial Intelligence Using Federated Learning���������������������������� 156

Manjushree Nayak and Debasish Padhi

Chapter 10 Applications of Federated Learning in AI, IoT, Healthcare,

Finance, Banking, and Cross-Domain Learning����������������������������� 175
Walaa Hassan and Habiba Mohamed

Chapter 11 Exploring Future Trends and Emerging Applications: A

Glimpse Into Tomorrow’s Landscape���������������������������������������������� 196
Utpal Ghosh and Shrabanti Kundu

Chapter 12 Securing Federated Deep Learning: Privacy Risks and

Countermeasures������������������������������������������������������������������������������ 228
Atharva Saraf, Shaurya Sameer Talewar, Susanta Das,
Khushbu Trivedi, and Ahmed A. Elngar

Chapter 13 IoT Networks: Integrated Learning for Privacy-Preserving

Machine Learning���������������������������������������������������������������������������� 250
Khushwant Singh, Mohit Yadav, Yudhvir Singh,
Pratap Singh Malik, Vikas Siwach, Daksh Khurana,
Binesh Kumar, Ramesh Kumar Yadav, and Ahmed A. Elngar

Chapter 14 Federated Query Processing for Data Integration Using

Semantic Web Technologies: A Review������������������������������������������� 276
Nidhi Gupta, Pawan Verma, Monali Gulhane, Nitin Rakesh,
and Ahmed A. Elngar

Index���������������������������������������������������������������������������������������������������������������������� 293
Preface
ARTIFICIAL INTELLIGENCE USING FEDERATED LEARNING:
FUNDAMENTALS, CHALLENGES, AND APPLICATIONS
The rapid evolution of artificial intelligence (AI) has reshaped various aspects of our
world, from healthcare to finance, and transportation to communication. Among the
latest advancements, federated learning (FL) stands out as a revolutionary paradigm,
enabling decentralized AI model training across multiple devices or organizations
while preserving data privacy and security. This book, Artificial Intelligence Using
Federated Learning: Fundamentals, Challenges, and Applications, aims to provide
a comprehensive understanding of this cutting-edge technology, bridging the gap
between theoretical foundations and practical implementations.
The inspiration for this book arises from the increasing need for privacy-pre-
serving AI solutions in an era marked by growing concerns over data security and
confidentiality. Traditional centralized AI models often require extensive data aggre-
gation, posing significant risks of data breaches and privacy violations. Federated
learning addresses these challenges by allowing models to be trained locally on edge
devices, sharing only the necessary updates to a central server, thus ensuring that
raw data remains secure and private.
This book is structured to cater to a diverse audience, including researchers,
practitioners, and students. It is divided into three main sections:

1. Fundamentals: This section introduces the core concepts of AI and feder-

ated learning. We delve into the theoretical underpinnings of FL, explain-
ing its architecture, key algorithms, and protocols. Readers will gain a
solid foundation in the principles that make federated learning a viable and
attractive approach to modern AI development.
2. Challenges: Despite its promising potential, federated learning faces sev-
eral challenges that need to be addressed to fully harness its capabilities.
This section explores issues such as data heterogeneity, communication
efficiency, model accuracy, and security concerns. We discuss ongoing
research and emerging solutions aimed at overcoming these obstacles, pro-
viding insights into the future directions of federated learning.
3. Applications: The final section showcases real-world applications of feder-
ated learning across various domains. Case studies and practical examples
illustrate how FL is being utilized to solve complex problems in healthcare,
finance, IoT, and beyond. Readers will learn about the benefits and limita-
tions of FL in practice, along with strategies for successful implementation.

As the field of federated learning continues to evolve, this book serves as both a
foundational text and a reference for advanced study. We hope to inspire and equip
our readers with the knowledge and tools necessary to contribute to this exciting and
impactful area of AI research and application.

vii
viiiPreface

We would like to express our gratitude to the many researchers and practitioners
whose pioneering work in federated learning has paved the way for this book. Their
contributions have been invaluable in shaping the content and direction of this work.
We also extend our thanks to our colleagues, reviewers, and readers for their support
and feedback.
We invite you to embark on this journey into the world of artificial intelligence
and federated learning, exploring the fundamentals, confronting the challenges, and
discovering the vast potential applications. May this book serve as a guide and inspi-
ration in your endeavors within this dynamic and transformative field.
About the Editors
Dr. Ahmed A. Elngar is an associate professor and Head
of the Computer Science Department at the Faculty of
Computers and Artificial Intelligence, Beni-Suef University,
Egypt. Dr. Elngar is also an associate professor of computer
science at the College of Computer Information Technology,
American University in the Emirates, United Arab Emirates.
Also, Dr. AE is Adjunct Professor at School of Technology,
Woxsen University, India. Dr. AE is the Founder and Head of
the Scientific Innovation Research Group (SIRG). Dr. AE is a
director of the Technological and Informatics Studies Center
(TISC), Faculty of Computers and Artificial Intelligence,
Beni-Suef University. Dr. AE has more than 150 scientific research papers published
in prestigious international journals and over 35 books covering such diverse topics
as data mining, intelligent systems, social networks, and smart environment. Dr. AE
is a collaborative researcher. He is a member of the Egyptian Mathematical Society
(EMS) and International Rough Set Society (IRSS). His other research areas include
the Internet of Things (IoT), network security, intrusion detection, machine learn-
ing, data mining, and artificial intelligence. big data, authentication, cryptology,
healthcare systems, and automation systems. He is an editor and reviewer of many
international journals around the world. Dr. AE has won several awards, including
the Young Researcher in Computer Science Engineering from the Global Outreach
Education Summit and Awards 2019, as well as the Best Young Researcher Award
(Male) (Below 40 years), Global Education and Corporate Leadership Award.

Dr. Diego Oliva In 2007, he obtained an electronics and

computer engineering degree from the Centro de Enseñanza
Técnica Industrial (CETI), the Industrial Technical Education
Center (CETI) of Guadalajara, Mexico, and an MSc in elec-
tronic engineering and computer sciences from the Universidad
de Guadalajara, Mexico in 2010. In 2015, he obtained a PhD
in informatics from the Universidad Complutense de Madrid
(UCM) in Spain. Since 2008, he has focused his research on
developing, implementing, and designing metaheuristic algo-
rithms. He has published more than 100 papers in international
journals on topics related to optimization and its implementa-
tions. Since 2015, he has been a member of the Academia Mexicana de Computacion
(AMEXCOMP), and since 2022, he has been a senior member of the Institute of
Electrical and Electronics Engineers (IEEE). Since 2017, he has been a member of
the Sistema Nacional de Investigadoras e Investigadores (SNII) in Mexico. In 2022,
he obtained the distinction of Highly Cited Researcher by Clarivate-Web of Science.
He is among the 2% most influential researchers worldwide, according to a report

ix
x About the Editors

published by Stanford University and Elsevier in 2023. He has been the editor and
author of several books at international publishing houses, and he is the associate
editor and guest editor for several specialized journals with high impact factors. He
is currently a professor and researcher at the Universidad de Guadalajara (CUCEI).
He also collaborates with Mexican and foreign universities in several research proj-
ects. His main research interests are artificial intelligence, metaheuristic optimiza-
tion algorithms, multiobjective optimization, parameter estimation in engineering,
and image and digital signal processing.

Prof. Valentina E. Balas is currently Full Professor in the

Department of Automatics and Applied Software at the
Faculty of Engineering, “Aurel Vlaicu” University of Arad,
Romania. She holds a PhD in applied electronics and tele-
communications from Polytechnic University of Timisoara.
Dr. Balas is the author of more than 300 research papers in
refereed journals and international conferences. Her research
interests are in intelligent systems, fuzzy control, soft comput-
ing, smart sensors, information fusion, modeling, and simula-
tion. She is the Editor-in Chief of the International Journal of
Advanced Intelligence Paradigms (IJAIP) and International
Journal of Computational Systems Engineering (IJCSysE), member of the editorial
board and member of several national and international journals, and an evaluator
expert for national and international projects and PhD theses. Dr. Balas is the direc-
tor of the Intelligent Systems Research Centre in Aurel Vlaicu University of Arad
and Director of the Department of International Relations, Programs and Projects
at the same university. She served as General Chair of the International Workshop
Soft Computing and Applications (SOFA) in eight editions, 2005–2018, held in
Romania and Hungary. Dr. Balas participated in many international conferences as
organizer, honorary chair, session chair and member in steering, advisory or inter-
national program committees. She is a member of EUSFLAT and SIAM; a senior
member of IEEE, and a member in TC–Fuzzy Systems (IEEE CIS), TC–Emergent
Technologies (IEEE CIS), and TC–Soft Computing (IEEE SMCS). Dr. Balas was
past Vice-President (Awards) of the IFSA International Fuzzy Systems Association
Council (2013–2015) and is a joint secretary of the Governing Council of the Forum
for Interdisciplinary Mathematics (FIM), A Multidisciplinary Academic Body,
India. She is also director of the Department of International Relations, Programs
and Projects and head of the Intelligent Systems Research Centre at Aurel Vlaicu
University in Arad, Romania.
Contributors
Valli Bhasha Achukatla Monali Gulhane
Electronics and Communication Symbiosis Institute of Technology
Engineering Department Symbiosis International (Deemed
KSRM College of Engineering University)
Kadapa, Andhra Pradesh, India Pune, Maharashtra, India

Muhammad Mujtaba Asad Nidhi Gupta

Sukkur IBA University Department of Computer Science and
Sukkur, Pakistan Engineering
School of Engineering and Technology
Luís Cardoso Sharda University
Polytechnic Institute of Portalegre and Greater Noida, Uttar Pradesh, India
Centre for Comparative Studies of
the University of Lisbon Walaa Hassan
Lisbon, Portugal Faculty of Computers & Informatics
Suez Canal University
Susanta Das Ismailia, Egypt
Ajeenkya D Y Patil University Faculty of Computer Science
Pune, Maharashtra, India Misr International University
Cairo, Egypt
Jagjit Singh Dhatterwal
Department of Artificial Intelligence Anupriya Jain
and Data Science Professor School of Computer
Koneru Lakshmaiah Education Applications MRIIRS Faridabad
Foundation New Delhi, New Delhi, India
Vaddeswaram, Andhra Pradesh, India
Vamsi Krishna Kadiri
Abdelaziz El-hoshoudy Electronics and Communication
Egyptian Petroleum Research Institute Engineering Department
Nasr city, Cairo, Egypt KSRM College of Engineering
Kadapa, Andhra Pradesh, India
Ahmed A. Elngar
Faculty of Computers and Artificial Balaram Yadav Kasula
Intelligence University of The Cumberlands
Beni-Suef University Williamsburg, Kentucky, USA
Beni-Suef City, Egypt
Kuldeep Singh Kaswan
Utpal Ghosh School of Computer Science and
Department of Computer Science Engineering
Sarojini Naidu College for Women Galgotias University
Kolkata, West Bengal, India Greater Noida, India

xi
xiiContributors

Pavan Kumar Kattela Appalaraju Muralidhar

Electronics and Communication Assistant Professor
Engineering Department Computer Science Engineering
KSRM College of Engineering VIT
Kadapa, Andhra Pradesh, India Chennai, India

Daksh Khurana Thangavel Murugan

Department of Computer Science & Assistant Professor
Engineering Department of Information Systems and
Symbiosis Institute of Technology Security
(SIT) College of Information Technology
Pune, Maharashtra, India University of United Arab Emirates
Al Ain, United Arab Emirates
Binesh Kumar
Research Scholar Muthmainnah
Department of Chemistry Universitas Al Asyariah Mandar
Guru Jambheshwar University of Polewali Mandar, Sulawesi Barat,
Science and Technology Indonesia
Haryana, India
Manjushree Nayak
Shrabanti Kundu Associate Professor
Department of Computer Science and Department of Computer Science and
Engineering Engineering
University of Kalyani NIST University
Kalyani, West Bengal, India Berhampur, Odisha, India

Kiran Malik Akkarapon Nuemaihom

Department of Computer Science and Buriram Rajabhat University
Engineering Buriram, Buriram, Thailand
Matu Ram Institute of Engineering and
Management Ahmad J. Obaid
Rohtak, Haryana, India University of Kufa
Najaf, Iraq
Pratap Singh Malik
Assistant Professor Debasish Padhi
Computer Science and Engineering Department of Computer Science and
Guru Jambheshwar University of Engineering
Science and Technology NIST University
Haryana, India Berhampur, Odisha, India

Ali Said Al Matari Nitin Rakesh

A’Sharqiyah University Symbiosis Institute of Technology
Oman Symbiosis International (Deemed
IAIN Curup University)
Sumatera, Indonesia Pune, Maharashtra, India
Contributors xiii

Rathipriya Ramalingam Yudhvir Singh

Assistant Professor Professor
Department of Computer Science Department of Computer Science &
Periyar University Engineering
Salem, India University Institute of Engineering &
Technology
Raveendra Reddy Maharshi Dayanand University
Assistant Professor Haryana, India
Computer Science Engineering
Vignan University Vikas Siwach
Guntur, India Assistant Professor
Computer Science and Engineering
Habiba Mohamed UIET, MDU
Faculty of Computer Science Rohtak, India
Misr International University
Cairo, Egypt Shaurya Sameer Talewar
Ajeenkya D Y Patil University
Atharva Saraf Pune, Maharashtra, India
Ajeenkya D Y Patil University
Pune, Maharashtra, India Khushbu Trivedi
Ajeenkya D Y Patil University
Prodhan Mahbub Ibna Seraj Pune, Maharashtra, India
American International University
Bangladesh (AIUB) M. Venkatanarayana
Dhaka, Dnaka, Bangladesh Department of Electronics and
Communication Engineering
Sivabalan Settu KSRM College of Engineering
Assistant Professor Kadapa, Andhra Pradesh, India
Computer Science Engineering
Vignan University Pawan Verma
Guntur, India Department of Computer Science and
Engineering
Seema Sharma School of Engineering and
Professor Technology
School of Computer Applications Sharda University
MRIIRS Greater Noida, Uttar Pradesh,
New Delhi, New Delhi, India India

Khushwant Singh Arkas Viddy

Research Scholar Universitas Al Asyariah Mandar
Department of Computer Science and Politeknik Negeri Samarinda,
Engineering Kalimantan, Indonesia
University Institute of Engineering and
Technology Idi Warsah
Maharshi Dayanand University Institut Agama Islam Negeri Curup
Haryana, India Curup, Sumatera, Indonesia
xivContributors

Pawan Whig Ahmad Al Yakin

Vivekananda Institute of Professional Universitas Al Asyariah Mandar
Studies-TC Polewali Mandar, Sulawesi Barat,
New Delhi, New Delhi, India Indonesia

Mohit Yadav Nikhitha Yathiraju

Assistant Professor University of The Cumberlands
Department of Mathematics Williamsburg, Kentucky, USA
University Institute of Sciences
Chandigarh University Syed Zahiruddin
Mohali, India Department of Electronics and
Communication Engineering
Ramesh Kumar Yadav KSRM College of Engineering
Assistant Professor Kadapa, Andhra Pradesh, India
Department of Data Science
Christ University
Bengaluru, India
 1 Overview, Challenges, and
Federated Learning

Ethical Considerations
Jagjit Singh Dhatterwal, Kiran Malik, Kuldeep
Singh Kaswan, and Ahmed A. Elngar

INTRODUCTION TO MACHINE LEARNING

Machine learning (ML) is a subfield of artificial intelligence. The objective of ML
is to exhibit to PCs new errands by breaking down existing information; along these
lines, they might make forecasts or assessments all alone, with next to no assistance
from people. The expression “AI” alludes to an umbrella term for different ways to
deal with information investigation, design identification, and knowledge creation
(Mitchell, 1997). Konečný et al. (2016) state that concentrated models are gener-
ally utilized by conventional AI draws near. To prepare and fabricate these mod-
els, conglomerating information from many sources and storing it in a focal area is
fundamental.
The most common way of preparing models utilizing information gathered from
a few hubs or sources and shipped off a solitary server is known as incorporated
learning (Yang et al., 2019). Many individuals are stressed over this methodology,
particularly about information security. The concentrated stockpiling of touchy
information improves the probability of safety breaks. Also, while working with
colossal data sets or information spread across numerous areas, brought together
advancing frequently faces adaptability issues (Kairouz et al., 2019). Incorporated
learning faces this impediment.
An option in contrast to the limits of unified learning has arisen with the com-
ing of united learning (FL). This objective is achieved by remembering cooperative
model preparation for scattered gadgets, as expressed by Bonawitz et al. (2017). The
exploration of Yang et al. (2019) recommends that to make Combined Learning (FL)
work better, information should be put away on nearby gadgets like cell phones, IoT
gadgets, or edge servers. This strategy permits these contraptions to learn together
while yet safeguarding buyer information. In contrast with unified techniques,
decentralization eases tension in information transport networks while safeguarding
the protection of clients’ singular data (Kairouz et al., 2019).
In combined learning, client gadgets do calculations on neighborhood informa-
tion and afterward send any refreshed models to a focal server or aggregator, as
expressed by Bonawitz et al. (2017). Yang et al. incorporate these progressions into a

DOI: 10.1201/9781003482000-1 1
2 Artificial Intelligence Using Federated Learning

total model that shields client protection while gathering information from different
sources in their 2019 exploration.

LITERATURE SURVEY
Unified Learning (FL) permits medical services associations to prepare models
cooperatively using delicate clinical information while safeguarding patients’ pro-
tection to an outrageous degree (Sheller et al., 2020). Florida (FL) has a few pur-
poses in various fields, one of which is medication. Monetary associations have
likewise researched information-secure forecast models utilizing united learning
approaches. Sending crude information to unified servers in an IoT ecosystem isn’t
needed; edge gadgets might gain from another and make adjustments depending
on the situation. Unified Learning (FL), first recommended in 2019 by Liu et al.,
makes this practical.
It is standard practice to utilize strategies like differential classification all
through the learning stage to keep the model precise while safeguarding the protec-
tion of individual data sources (Shokri et al., 2015). The field of Unified Learning
(FL) keeps on focusing on the assurance of clients’ very own data. Moreover, uni-
fied learning (FL) frameworks frequently utilize safe conglomeration systems and
encryption strategies to ensure the mystery of model changes during transmission
(Bonawitz et al., 2017).
Similarly, as with each innovation, FL has its extraordinary arrangement of
issues. As per concentrates finished by Li et al. in 2020, gadget heterogeneity is a
significant impediment to arriving at an overall model union. This term suggests
aberrations in information conveyance and handling power between gadgets. As
per Hard et al., unified learning frameworks may not work as proficiently when
confronted with correspondence issues like drowsy associations or confined trans-
fer speed.
Konečný et al. (2016) state that there is constant work to further develop calcula-
tion correspondence proficiency, reinforce models to endure antagonistic attacks,
and take care of the issues of security precision balance. There is a consistent expan-
sion in the field’s leap forward, and progressing research in unified learning is cur-
rently centred around specific points. Research on united strategies for advancement
that permit contribution in decentralized learning while at the same time holding
model proficiency is presently in progress by Smith et al. (2021).

UNDERSTANDING FEDERATED LEARNING

An inventive way to deal with AI is unified learning, which takes into account the
solid and scattered creation of models on different edge devices (Yang et al., 2019).
Combined learning is another method that is just entering the AI crowd. Devices
such as cell phones, Web of Things (IoT) devices, and edge servers are working
together to create a joint model without moving natural data (Kairouz et al., 2019).
Coordinated efforts and shared information are the foundations of the decentral-
ized learning worldview. The creation of AI models using decentralized information
Federated Learning 3

sources that guarantee information obfuscation was proposed by Google in a signifi-

cant report in 2017 (Bonawitz et al., 2017).
This effort is responsible for the advancement of unified learning. The inherent
advantages of this field have led to a huge expansion of innovative work aimed at
exploring its potential uses in many areas (Kaswan et al., 2022a).
According to Li et al. (2020), the core idea of unified learning is to decentralize
information processing so that models can be created on any teaming-up gadget
or hub. According to Mohassel and Zhang (2017), privacy and security concerns
are sufficiently addressed by storing sensitive customer information on the devices
that create it instead of sending it to a central server. Konečný et al. (2016) state
that new AI computations, strategies for mediation, and security-saving frameworks
have contributed to the further development of combined learning. By integrating
unified learning into a grounded system, this development effectively nullifies the
disadvantages of concentrated learning. Unified learning is a significant advantage
in situations where the protection of sensitive data is essential. Information breaches
and unlawful access to individual data are less likely with this approach, as Yang
et al. (2019) show.
United Learning smoothes the cycle by allowing models to be created at unique
points. According to Hard et al. (2018), this strategy improves the performance and
adaptability of AI models while simplifying the handling of different data sets dis-
tributed across multiple devices. The problems caused by information storage and
policies could be better solved with the help of combined learning.
Table 1.1 Shows the summary of the federated Learning.

TABLE 1.1
History of Federated Learning
Year Milestones and Key Developments in Federated Learning
2016 Google Brain introduces federated learning, aiming to enable on-device machine learning.
2017 Initial research publications on federated learning emphasize privacy-preserving aspects.
2018 Advancements in differential privacy techniques for federated learning emerge in
research.
2019 Google releases federated learning Frameworks to facilitate decentralized model training.
2020 Exploration of secure and communication-efficient protocols for federated learning
begins.
2021 Federated Learning applications expand into healthcare, IoT, finance, and
telecommunications.
2022 Ongoing research focuses on federated transfer learning and lifelong federated learning.
2023 Innovations in federated meta-learning and adaptive aggregation techniques gain attention.
2024 Integration of federated learning with edge computing and federated reinforcement
learning.
Present Continued advancements in privacy-preserving methods and broader adoption in various
domains.
4 Artificial Intelligence Using Federated Learning

KEY COMPONENTS OF FEDERATED LEARNING

Recently, a creative and free strategy for learning and teaching has developed.
Achieving the ideal outcome depends on the targeted activities of some important
elements (Kaswan et al., 2022b). At the center of consideration are client devices
that often use edge registration. These devices can perform computations while in
proximity. A wide range of electronic devices can utilize their information to cre-
ate bounded models; this includes everything from cell phone information to IoT
reviews. This technique limits the ability to share information with other devices.
As described by Yang et al. (2019), each device is part of the organization and main-
tains its model nearby by sending adjustments to the central server, which in some
cases is called an aggregator. The process of model accumulation is controlled by
the aggregator, which can be set up on-premise or in the cloud. For this purpose, the
framework collects and combines updates from numerous client devices. Bonawitz
et al. (2020) refer to a combination of authors.
To guarantee a protected and smooth connection between the client’s hardware
and the specialist, Unified Learning utilizes the best correspondence advancements
and security strategies. These innovations not only facilitate the exchange of model
limits and updates, but likewise ensure the security of client information and data
(Dhatterwal et al., 2022a). It is fundamental to use encryption techniques such as
differential security and the interplay of learning and explicit secure accumulation to
ensure the protection of information during model extraction. McMahan, H. B. and
others authored the 2017 paper. To guarantee that nearby models match the focal PC
and to diminish the probability of information mistakes, unified averaging is a great
approach for correspondence security. According to Konečný et al. (2016), it is criti-
cal for the creation of neighborhood models that edge client gadgets have reasonable
computational and capacity limits.
The research by Nishio and Yonetani designs an edge-driven method that effec-
tively reduces idle state by reducing the repetition of information transmission to
a focal server. By storing the basic data on the device, this system also increases
security (Kaswan et al., 2021). All model updates from the devices can be merged
and collected on the focal server. It also acts as an operational hub for collecting
information and creating thoughts. As Yang et al. (2019) point out, it is possible to
create a global model by merging knowledge data from different provincial data sets
into a common data set.
To function appropriately, Unified Learning relies on several correspondence
components. The business offers arrangements that enable client gadgets and the
focal specialist to exchange information and synchronize models. The protection of
information originating from individual devices can be ensured through the use of
a shared secure total. This ensures that changes to the model can be incorporated
securely. The 2017 paper was authored by McMahan, H. B., et al. Protecting cli-
ents’ information gathering by concealing their data is one of the many correspon-
dence-related uses for differential security measures. According to Bonawitz et al.
(2020), the research refers to various authors. The security and well-being of the
Unified Educational Experience is ensured through the execution of these aggregate
safeguards.
Federated Learning 5

HOW FEDERATED LEARNING WORKS

Helpful preparation of models across multiple edge gadgets or clients is enabled
by Combined Averaging Calculation, a decentralized rationalization approach.
For Combined Figuring out how to work, this system should be followed. In this
technique, the model is refreshed locally before it is sent to a focal server for
shuffling, which utilizes the neighborhood information to prepare the gadgets.
Specialists found that the gadgets involved in the review made changes after get-
ting the accumulated model and sending it back. The secrecy of the information
is ensured by the self-reproducing learning system of this system (Dhatterwal
et al., 2022).
Numerous means are utilized in decentralized model preparation. Deciding
which gadgets are useful is the most important component. Each gadget puts away a
specific data set. Rest assured that all data stored on these gadgets is kept secret, even
if they could benefit from their information. The preferences of the learned models
are sent after preparing interaction with the fundamental PC. It is the duty of the
focal server to ensure that all gadgets update and gather these elements at the same
time (Yang et al., 2019).
To create a global model, Unified Gaining uses the pooling technique to merge
data from different devices. The use of a weighted typical method is common. The
system integrates the device feedback, the objective result of the model, and other
indicated measurements into the boundaries of the model.
The synchronization of subsequently accumulated information across all gadgets
is important for the model refresh cycle in United Learning. From the main PC,
each gadget in the process receives the latest adaptation of the worldwide model.
The devices can then utilize the acquired information to work on their neighborhood
models. As described by Kairouz et al. (2019), the iterative system plans to improve
the execution and understanding of the model through repeated patterns of model
selection, adaptation, and rearrangement.
The Unified Averaging Calculation tracks down the world’s best arrangement by
merging appropriate model updates with neighborhood calculations. As outlined in
the 2020 study by Li et al., the understanding shown is achieved after a comprehen-
sive conversation and the formation of a point-by-point model. Through this process,
people can participate in cooperative learning while ensuring the classification of
information collected from each device.
It is mandatory to create models openly and straightforwardly to reduce secu-
rity issues related to the consistent maintenance and handling of information.
Consequently, sensitive customer data is not yet stored on the devices in the customer
area.
The acquisition systems utilized in unified learning influence the viability and
exactness of the worldwide model. Exploring different acquisition methods should
reduce biases and ensure a fair evaluation of the contributions of different devices.
Next, some models are introduced: fundamental averaging, weighted averaging, and
discrimination-based collection of protection data.
United Learning uses model updating methods to ensure that model boundaries
remain reliable across multiple devices, regardless of whether those devices have
6 Artificial Intelligence Using Federated Learning

different information boundaries and processing capacities. Unified Averaging

with Force (FedAvgM) and Government Stochastic Angle Plunge (FedSGD)
are two strategies used in appropriate detection to increase combination and
robustness.
The central PC in Unified Advancing not only collects information but also man-
ages various other functions. It controls the organizational connections of devices,
monitors the progress of model updates, and ensures that the global model is accu-
rate and predictable. As Yang et al. (2022) point out, the server simplifies the execu-
tion of United Advancing by organizing the compilation and appropriation of models
for each of the affected devices.
Since combined learning is iterative, it includes some patterns of nearby prepara-
tion and worldwide gathering. As a result, the representation of the model can get
better and better. According to Konečný et al. (2016), the worldwide model works
on its representation with each cycle by utilizing different sources of information
and consolidating the collected knowledge while ensuring the protection of the
information.
Two protection techniques, secure collection, and differential security are stated
by United Figuring to guarantee that delicate information stays secret throughout
the model conglomeration. According to Mohassel and Zhang (2017), these strate-
gies ensure that the merged model adheres to all guidelines for protecting individu-
als’ data.
Due to its characteristic freedom, Combined Learning can continue with the tasks
regardless of whether there are difficulties with the network or defective hardware.
The collaborative cycle will continue in any case, even if devices fail or companies
experience delays, as a 2020 study by Smith et al. shows. To achieve this goal, it
could be very useful to include more real devices in the model accumulation process.
Flexibility and resilience to disappointment are certainly of great importance in this
context.
United Learning can effectively solve issues brought on by the absence of free-
dom and personality confirmation in information, changes in gadget types, and
heterogeneous information circulation by utilizing enhancement options. Weighted
inspection, customizable learning rates, and model transformation are among the
techniques recommended by Li et al. (2020) to mitigate the effects of these prob-
lems. At the end of the day, these strategies lead to an easier demonstration of blend-
ing and a more generally talked about execution.
The features and pace of unification of unified learning procedures are the cur-
rent focus of assembly research. The examination by Dinh et al. (2020) gives insight
into how far the gadgets are occupied with cooperative improvement. The outcomes
demonstrate that the technique is exact and dependable.
Unified learning has long struggled with the problem of productively dealing with
the proliferation of non-ID information. Some research focuses on effectively address-
ing cooperative learning calculations that can deal with non-delegated information
designs across different devices to address this issue. Yang et al. (2022) found that these
procedures act on the trustworthiness and generalizability of the prepared models.
Federated Learning 7

It is undoubtedly conceivable to circumvent the limits of performance and speed

that can be encountered when utilizing combined learning in edge-figure environ-
ments. For the advancement of edge insights, combined learning is fundamental to
streamline model preparation and updating on low-capital devices (Gao et al., 2021).
We focus on expanding the utilization of existing PC and transmission innovations
to achieve this goal.
To address the security issues with United Learning, we are looking for areas of
strength. The utilization of harmful strategies in models and the portrayal of byz-
antine practices are two conspicuous examples of these limitations in practice. To
make United Learning frameworks more secure, specialists are examining various
methods. The aforementioned parts incorporate conventions for safe amalgamations,
calculations that can survive Byzantine shortcomings, and differential protection-
based shields.
The Combined Learning area is currently researching novel methods such as ver-
satile collection and unified meta-learning. The goal is to improve model combina-
tion and execution while dealing with different gadgets and non-IID information.
The goal of these complex pooling approaches is to work on cooperative learning in
cluttered, information-rich environments.
Unified supportive learning, constant combined learning, and unified mov-
ing learning are dynamic areas of research. The commitment to collabora-
tive learning is rapidly advancing the progress of these projects. According to
researchers, the goal of these updates is to make cooperative methods relevant
to a broader range of learning circumstances and certifiable challenges (e.g.,
Chen et al., 2023).

The stages involved in a typical FL process as shown in Figure 1.1:

• Initialization: The allocation of a centrally run first global model with ran-
dom, initial weights is done by a central server.
• Client Selection: The server chooses a group of available equipment (par-
ticipants) to be in the learning round among the various equipment that
present themselves.
• Model Distribution: The server delivers the current global model’s weight
to clients which was the one chosen.
• Local Training: The client will proceed to perform the training using their
local data and subsequently get the updated model weights.
• Model Updates: Through sending once-updated model weights to our
server, clients perform this function.
• Aggregation: The server’s job is to combine these updates normally, by
averaging, to correct the global model.
• Model Evaluation: The server uses the global model to appraise it to figure
out its efficiency.
• Iteration: After steps 2 to 7 are performed, the process is repeated until the
model converges or the desired performance state is attained.
8 Artificial Intelligence Using Federated Learning

FIGURE 1.1 federated learning Algorithm

CHALLENGES AND LIMITATIONS

The security of information is crucial in the field of combined learning, as the infor-
mation is only processed in a few places. It is essential to develop and apply sound
protection strategies while acquiring and maintaining models to reduce the likeli-
hood of information breaches and security violations.
The extra effort required to design connections is a major issue that arises when
multiple devices participate in correspondence. Expanded correspondence costs and
potential breaks in fixing issues could result from the standard exchange of model
changes between the main PC and other devices. This could affect the student’s data
security, as indicated by an examination.
Federated Learning 9

In situations where there are not many devices or the networks are not suitable for
handling many information movements, there may be capacity issues in data trans-
fer when transferring model updates between devices. According to Konečný et al.
(2016), managing models with enormous scale or a large number of devices causes
ongoing processes to stall. As a result, model synchronization becomes less produc-
tive and reliable, which increases the time required for interaction.
Since combined learning uses a wide range of devices and information sources,
the execution is more complicated. Differences in processing capacity, informa-
tion dispersion, and the steadfastness of gadgets could make it a difficult endeavor
to coordinate information from many sources while keeping up the precision and
impartiality of the prepared model (Karimireddy et al., 2021).
Regardless of whether encryption and other measures are utilized to improve
protection in general, concerns about security and information insurance continue to
emerge. Xie et al. (2023) emphasize the necessity of great strength areas for execut-
ing systems and secure association methods to shield the united profits from mali-
cious endeavors like model harm or supposition assaults.
It is fundamental to execute model printing strategies and advance correspon-
dence conventions to diminish information transfer rates between gadgets and the
main PC. It is fundamental that you think about the transmissions by performing the
expected steps. Quantization and differential pressure are two of the techniques used
to enhance the cost-effectiveness of correspondence without compromising model
execution).
It is fundamental to utilize imaginative common learning strategies that can
effectively oversee fluctuating information designs and unmistakable qualities since
gadgets and information indices are noteworthy by nature. Modeling the unification
cycle as well as execution in different environments could be improved by effectively
answering device factors through the use of versatile overall methods and custom-
ized combined learning calculations).

APPLICATIONS OF FEDERATED LEARNING

With the introduction of Combined Learning, an advanced strategy that enables the
collaborative preparation of models while effectively regulating protection issues,
medical care and testing in medicine has fundamentally changed. In addition, by
utilizing diverse patient data, medical institutions can gain significant experience
in protecting individual information about patients. By decentralizing the gathering
of information and simplifying the development of expectation models, this goal is
achievable.

Healthcare and Medical Research

For healthcare and clinical research organizations, this is a top priority since it
enables collaborative AI model training without compromising patient privacy. This
technology makes it possible for medical organizations, research centers, and phar-
maceutical companies to develop AI models jointly so that the privacy of the data is
10 Artificial Intelligence Using Federated Learning

not compromised. With federated learning, models can access and learn from dis-
tributed data sets which are dispersed across multiple healthcare institutions while
data security is preserved (Sheller et al., 2020). The use of this method can be seen
in the improvement of disease detection and clinical imaging. Furthermore, organi-
zations can follow strict privacy rules such as GDPR or HIPAA through federated
learning where it is possible to just utilize useful insights from diverse data sets
without directly sharing those sets of data (Yu et al., 2021). This enables the pharma-
ceutical people to run trials effectively.

Internet of Things (IoT) and Smart Devices

Federated Learning will be able to achieve the goals of the IoT ecosystem by perform-
ing complex algorithms on devices with limited resources. Several smart devices like
sensors, wearables, and smart appliances may not have sufficient computing power
and internet connection. By using the method of federated learning, models can be
trained across multiple devices without transferring raw data to centralized servers,
as reduced communication is a benefit of such an approach (Konečný et al., 2016).
The applications span various areas such as improvement of anomaly detection in
IoT systems, customization in smart homes, and enabling predictive maintenance in
industrial IoT setups (Lim et al., 2020).

Financial Services and Banking

The finance sector has been leading in predictive modelling, risk identification,
and fraud detection while taking into account the adoption of data protection
regulations, thereby ensuring compliance through the use of federated learning.
Banks and other financial institutions may protect their customers’ private data
with the help of federated learning methods. This is achieved through a distrib-
uted network of branches and ATMs to retrain the A1 models. The feature of
federated learning guarantees that customers preserve their privacy while infor-
mation required for loan approval and credit scoring processes is gathered from
different sources (Chen et al., 2019). The capability to facilitate collaboration
among financial institutions in sharing insights on fraudulent behavior patterns is
one of the advantages of adopting federated learning in fraud detection systems.
As reported by Zhang et al., the cooperative way guarantees the privacy of highly
confidential transactional data.

FEDERATED LEARNING IN INDUSTRY AND RESEARCH

Among the areas where federated learning has been applied widely are healthcare,
telecommunications, and finance. It can give healthcare institutions the indepen-
dence to create collaborative disease prediction models without depending on cen-
tralized data collection.
Telecommunication firms can improve network speed and customer satisfaction
by deploying federated learning. This is done by jointly training models in this net-
work that distribute the base stations. Gupta et al. raise the point that this method
Federated Learning 11

is useful in several areas such as resource allocation, spectrum management, and

predictive maintenance which reduces data transmission overhead.
Research on federated learning is a relatively new but quickly evolving area.
Scientists are involved in a quest for newly emerging technological applications,
optimization methods, and mechanisms for privacy protection.
Further research in federated learning should focus on its potential use for other
topics, including federated reinforcement learning, federated transfer learning, and
lifelong federated learning. The main purpose of Chen et al. (2023) is to develop a
federated learning technique that optimizes performance according to the different
learning strategies and difficult real-world situations. At present, research is focused
on advanced data patterns and techniques.

COMPARISON WITH OTHER LEARNING PARADIGMS

Centralized learning centralizes all data in one location for effective model training.
The main advantage is the access to a huge set of data, which in turn leads to a more
efficient model. On the other hand, the prevalence of data breaches and the violation
of privacy become more feasible as centralized data storage increases the chances of
privacy breaches.
Decentralized System Analysis: Distributed Learning provides model training
amongst several nodes or servers to eliminate the need for a central aggregator.
Single nodes work with smaller data components, while others coordinate to update
a global model. Despite this approach being a distributional one, the problem of
high communication costs has been caused by an excessive number of intra-node
connections.
Federated Learning: To assist the learning of new information, federated learning
takes into account the characteristics of both distributed and centralized learning
models. It keeps the data secure, while the models are trained on scattered devices.
This function lets one train their mode locally instead of transmitting the data to a
remote server eliminating any security issues. Through the process of updating the
model parameters but not the raw data, federated learning is suitable for privacy-
sensitive applications.
Through decentralization, federated learning limits the susceptibility to data
breaches. User privacy is kept as sensitive data resides on local devices. As a result
of decentralized learning, there arises no need for centralized data distribution. The
importance of this approach is twofold: firstly, it helps collaborative model learning
by leveraging multiple devices with heterogeneous data sets. Through the use of
distributed computing, federated learning takes advantage of multiple computing
devices which in turn, leads to efficiency in processing while minimizing the costs
of data transmission.

ETHICAL AND LEGAL CONSIDERATIONS

Training AI models based on distributed data across multiple devices is what fed-
erated learning does. Since data usually remains on consumer devices for a long
time, it is difficult to tell the ownership of the aggregated model, which results in
12 Artificial Intelligence Using Federated Learning

data ownership complications. The rights and responsibilities must be set as clear
as possible among data contributors, model developers, and aggregators to achieve
peace and the shared utilization of the data provided (Bagdasaryan et al., 2020). The
consent of the users is critical in the case of Federated learning as well as when the
data is retrieved from personal devices.
Users will be expected to have an extensive knowledge of data usage, model train-
ing objectives, and the problems of the federated learning projects before its full imple-
mentation. It is informed consent that allows users to make wise selections regarding
their data submission and promotes transparency. Unlike some of the privacy stan-
dards, such as the General Data Protection Regulation (GDPR) of the European Union,
the Health Insurance Portability and Accountability Act (HIPAA) of the healthcare
sector, and a variety of local laws, federated learning is not met by their norms inher-
ently. Data anonymization, encryption, and secure aggregation are some of the tech-
niques that will be required to protect user privacy. Yang et al. mention in their 2022
paper that it is vital to implement such measures to meet the requirements.
The field of federated learning is based on ethical behaviour. The concepts of
non-discrimination, transparency, accountability, should be at the core of everything
that follows before the start of federated learning, according to Yang et al. (2022).
Federated learning systems have to comply with existing regulatory settings. Naseri
et al. (2022) indicate that this approach should be based on privacy-by-design prin-
ciples, robust security measures, and the obtaining of the respective licenses or cer-
tificates. Bonawitz et al. (2020) suggested that federated learning systems need to
be constantly observed and modified to keep up with shifting privacy laws, ethical
standards, and legal requirements.

FUTURE DIRECTIONS AND EMERGING TRENDS

Strategies for Enhancing Privacy: As summarized by McMahan et al. (2018), robust
security measures have been brought about by combining powerful privacy-preserving
technique like homomorphic encryption, secure multi-party computation, and dif-
ferential privacy to the federated learning.
More refined aggregation algorithms would give rise to better uniformity, fair-
ness and adaptability to the different kinds of information and device capabilities.
Hence, aggregation methods such as hierarchical aggregation, adaptive aggregation,
and federated meta-learning have been suggested in the literature (Li et al., 2024).
Streamlined Communication Methods: Researchers currently work on the tech-
niques of federated learning, which should minimize communication load while
keeping model precision and convergence. Asynchronous communication protocols
and decentralized development strategies are included in these tactics (Yang et al.,
2023). The models are able to acquire new knowledge and the way they work can be
changed dynamically to adjust to different situations. Among others, lifelong feder-
ated learning and federated transfer learning are the two areas where this phenom-
enon is predominant. These approaches in fact support a better information flow
between devices, domains or tasks.
After the deployment of federated learning the process of training AI models will
be changed in a radical way. Part of this transition involves decentralization from
Federated Learning 13

centralized data repositories to distributed and privacy-preserving technologies. The

initiatives involved in raising the privacy and security standards might have a posi-
tive impact on subsequent advancements in federated learning. These regulations
protect the privacy of individuals and ensure ethical data management, which will
serve as the basis for both federated approaches and generalized artificial intelli-
gence methods.
Federated Learning’s Emphasis on Collaboration: The tendency towards collabo-
ration and open-source principles in federated learning can kindle a more coop-
erative attitude and data sharing among the AI developers, thereby bringing the
increased adoption of open-source tools and methodologies.

CONCLUSION
Federated Learning redesigns AI paradigms by combining data privacy with distrib-
uted model training. Thanks to decentralized data processing, it becomes possible to
train the models here collaboratively across different devices and still ensure private
information. The Federated Averaging Algorithm provides for the uncomplicated
integration of the device-specific features into a centrally updated global model
which also eliminates operational complexity. The various advantages of federated
learning indicate that it has a promising future. The deployment of edge devices, on
the other hand, can decrease the privacy issues of centralized systems. This paradigm
shift allows us to utilize the wide variety of data sources without the need to violate
any privacy laws. In effect, federated learning emerges as a very useful solution in
the healthcare and the Internet of Things fields, where resources are limited and data
is extremely sensitive. Having your data secured, you can go on to build your model
without any concern. With research, many new achievements for federated learning
are expected to be achieved in the future AI is about to witness a major change as
a result of improved privacy protection, well-established communication protocols,
and integration of today and tomorrow technologies such as Internet of things and
Edge computing. Now, a new era has begun where these emerging phenomena are
emphasized. This new era incorporates distributed model training, accountable data
management, and collective intelligence. Federated learning is an example of novel
AI solutions that are founded on privacy protection, collaborative sphere and always
in compliance with ethical principles in dealing with private data. It could have a
deep influence on how models are trained and used, and therefore, is critical in the
development of a reliable and ethically sound AI. Its ability to skillfully handle pri-
vacy concerns makes this philosophy differ from other AI ways.

REFERENCES
Bagdasaryan, E., Veit, A., Hua, Y., Estrin, D., & Shmatikov, V. (2020). How to backdoor
federated learning. arXiv preprint arXiv:2007.14191.
Bonawitz, K., Eichner, H., Grieskamp, W., Huba, D., Ingerman, A., Ivanov, V.,. . .& Moore, D.
(2017). Practical secure aggregation for privacy-preserving machine learning. In CCS ’17:
Proceedings of the 2017 ACM SIGSAC Conference on Computer and Communications
Security (CCS) (pp. 1175–1191). https://doi.org/10.1145/3133956.3133982
14 Artificial Intelligence Using Federated Learning

Bonawitz, K., Grieskamp, W., Huba, D., Ingerman, A., Ivanov, V., Kiddon, C.,. . .& Vasilaky,
K. (2020). Towards federated learning at scale: A systematic comparison of privacy-
preserving aggregation algorithms. arXiv preprint arXiv:2007.10987.
Chen, K., Lin, J., Yang, Q., Zhou, P., & Zheng, Y. (2023). Federated reinforcement learning:
A comprehensive review. arXiv preprint arXiv:2302.04567.
Chen, M., Xu, X., Zhang, Z., & Cheng, S. (2019). Federated learning in mobile edge networks:
A comprehensive survey. IEEE Communications Surveys & Tutorials, 21(3), 3262–3293.
Dhatterwal, J. S., Kaswan, K. S., Jaglan, V., & Vij, A. (2022a). Machine learning and deep
learning algorithms for IoD. In The Internet of Drones: AI Applications for Smart
Solutions (p. 237). CRC Press.
Dhatterwal, J. S., Kaswan, K. S., Preety, D., & Balusamy, B. (2022b). Emerging technologies
in the insurance market. In Big Data Analytics in the Insurance Market (pp. 275–286).
Emerald Insight.
Dinh, T., Tran, T., Pham, T., Lee, S., & Kim, Y. (2020). Convergence analysis of federated
learning: A survey. arXiv preprint arXiv:2005.07089.
Gao, L., Wu, Q., Wei, Z., Zhang, L., & Zhang, H. (2021). Federated learning for edge intel-
ligence: A comprehensive survey. IEEE Communications Surveys & Tutorials, 23(1).
Hard, A., Rao, K., Mathews, R., Ramaswamy, S., Beaufays, F., Augenstein, S.,. . .&
Eichner, H. (2018). Federated learning for mobile keyboard prediction. arXiv preprint
arXiv:1811.03604.
Kairouz, P., McMahan, H. B., Avent, B., Bellet, A., Bennis, M., Bhagoji, A. N.,. . .&
Ramage, D. (2019). Advances and open problems in federated learning. arXiv preprint
arXiv:1912.04977.
Karimireddy, S. P., Stich, S. U., Jaggi, M., & Hoffmann, M. (2021). Adaptive federated opti-
mization. arXiv preprint arXiv:2102.05203.
Kaswan, K. S., Dhatterwal, J. S., & Kumar, K. (2021). Blockchain of internet of things-based
earthquake alarming system in smart cities. In Integration and Implementation of the
Internet of Things Through Cloud Computing (pp. 272–287). IGI Global.
Kaswan, K. S., Dhatterwal, J. S., Kumar, S., & Lal, S. (2022b). Cybersecurity law-based
insurance market. In Big Data: A Game Changer for Insurance Industry (pp. 303–321).
Emerald Publishing Limited.
Kaswan, K. S., Dhatterwal, J. S., Sharma, H., & Sood, K. (2022a). Big data in insurance inno-
vation. In Big Data: A Game Changer for Insurance Industry (pp. 117–136). Emerald
Publishing Limited.
Konečný, J., McMahan, H. B., Ramage, D., & Richtárik, P. (2016). Federated optimization:
Distributed optimization beyond the datacenter. arXiv preprint arXiv:1511.03575.
Li, T., Sahu, A. K., Talwalkar, A., & Smith, V. (2020). Federated learning: Challenges, meth-
ods, and future directions. IEEE Signal Processing Magazine, 37(3), 50–60.
Lim, B., Kang, J., Jung, J. J., & Oh, S. (2020). Federated learning for Internet of Things:
A comprehensive survey. IEEE Access, 8, 23537–23565.
McMahan, H. B., Abadi, M., Chu, A., Goodfellow, I., McMahan, B., Mironov, I.,. . .&
Ramage, D. (2018). Advances and open problems in federated learning. arXiv preprint
arXiv:1812.09975.
Mitchell, T. (1997). Machine Learning. McGraw Hill.
Mohassel, P., & Zhang, Y. (2017). Secureml: A system for scalable privacy-preserving
machine learning. In 2017 IEEE Symposium on Security and Privacy (SP), San Jose,
CA, USA (pp. 19–38). IEEE Xplore. doi: 10.1109/SP.2017.12
Sheller, M. J., Edwards, B., Reina, G. A., Martin, J., Pati, S., Kotrotsou, A., & Milchenko, M.
(2020). Federated learning in medicine: Facilitating multi-institutional collaborations
without sharing patient data. Scientific Reports, 10(1), 1–11.
Federated Learning 15

Shokri, R., Shmatikov, V., & Smith, M. (2015). Privacy-preserving deep learning. In
Proceedings of the 22nd ACM SIGSAC Conference on Computer and Communications
Security (pp. 1310–1321). Association for Computing Machinery.
Smith, V., Chaudhuri, K., Sanjabi, M., & Talwalkar, A. (2021). Federated learning: Strategies
for improving communication efficiency. arXiv preprint arXiv:2106.03638.
Xie, H., Li, Y., Xu, T., & Zhang, Z. (2023). Secure and robust federated learning: A compre-
hensive survey. arXiv preprint arXiv:2305.01972.
Yang, Q., Liu, Y., Chen, T., & Tong, Y. (2019). Federated machine learning: Concept and
applications. ACM Transactions on Intelligent Systems and Technology (TIST), 10(2),
1–19.
Yang, Q., Liu, Y., Chen, T., & Tong, Y. (2022). Federated machine learning: Concept and
applications. ACM Transactions on Intelligent Systems and Technology (TIST), 11(2),
1–19.
Yang, Q., Liu, Y., Chen, T., & Tong, Y. (2023). Communication-efficient federated learning:
Recent advances and future directions. ACM Transactions on Intelligent Systems and
Technology (TIST), 10(2), 1–19.
Yu, K. H., Beam, A. L., & Kohane, I. S. (2021). Artificial intelligence in healthcare. Nature
Reviews Drug Discovery, 20(5), 373–374.
 2 In-Depth Analysis of
Artificial Intelligence
Practices
Robot Tutors and Federated
Learning Approach in
English Education
Muthmainnah, Akkarapon Nuemaihom,
Ahmad Al Yakin, Prodhan Mahbub Ibna Seraj,
Muhammad Mujtaba Asad, and Ahmed A. Elngar

INTRODUCTION
According to Khang et al. (2023a), integrating traditional education into complex
systems is gaining increasing attention because of the integration of artificial intel-
ligence (AI) with education. Advances in new education curricula and infrastruc-
ture systems have led to the widespread adoption of artificial intelligence in various
fields, including administration, resource development, and more. Research on intel-
ligent education has so far concentrated on methods to improve students’ learning
efficiency and intelligent education monitoring, but limited research has examined
how to assess the efficacy and level of teaching provided by teachers utilizing AI
robotics. Indeed, a variety of factors, such as students’ efforts, the availability of
contemporary learning tools, and the style of their instructors, affect their ability to
learn (Muthmainnah et al., 2023). We present a smartphone-based robotics learning
system based on blended learning for educators. By understanding the feelings of
undergraduate students when they learn a foreign language, the realization of feder-
ated learning (FL) improves recognition performance and enables data sharing and
collaborative modelling.
The new idea of Industry 6.0 highlights the importance of human workers in
production with increasingly harmonious technological collaboration. The goal of
Industry 6.0 is to combine the knowledge and experience of human adaptation with
advanced technologies to make production processes more adaptive and collabora-
tive (Chourasia et al., 2022). This 6.0 industrial revolution enables the sector to com-
bine human and robotic workers by adopting the former key enablers such as the

16 DOI: 10.1201/9781003482000-2
In-Depth Analysis of Artificial Intelligence Practices 17

Internet of Things (IoT), big data analytics, and artificial intelligence. The goal is to
have a highly collaborative, adaptable, and agile team (Khalid et al., 2023).
By these statements, the robot tutor that AI-driven technology is one of the
edge devices that can be used together with FL, where this collaboration empha-
sizes cooperation between human–robot interaction (HRI) to train and improve
new machine-based skills. Through this practice, the virtual robot learning model
adapts to the learning environment without having to send data and information to
a remote server, which poses a security risk for certain bot programs (Papadopoulos
et al., 2021). This FL approach includes safety measures and more efficient and
effective human–robot collaboration, which is supported by robot facilitation that is
consciously personalized by the surrounding environment (Lindblom et al., 2020).
Subsequently, this concept is adopted for teaching modern English and individual
production, where FL has an impact in the field of education at large. This statement
is supported by the results of research conducted by Sadiku et al. (2022) and Driss
et al. (2023), who suggest that the FL approach can help schools create AI-driven
learning models or machine learning by protecting user data. Protecting user data
is highly recommended so that the use of AI-driven learning applications can be
adjusted, and manufacturing becomes more personalized and user-centered by
utilizing insights gained by customizing specific products, processes, or services
such as AR, VR, MR, or other educational technologies (Shakeer & Babu, 2024;
Wu et al., 2023). Taking this route has the potential to increase the practicality and
efficacy of AI, augmented reality, and virtual reality experiences in educational or
teaching contexts by making them more immersive and realistic (de Moraes Rossetto
et al., 2023; Khang et al., 2023a; Klimova et al., 2023). Moreover, the FL approach
in English as a Foreign Language (EFL) can help students learn from each other and
share what they know.
Adopting the federated learning paradigm as outlined by Angurala and Khullar
(2023), our objective is to optimize English language teaching through decentralized
machine learning models. Previously, in traditional language classrooms, all data and
learning insights usually resided on a centralized server, like some previous learning
platforms. In federated learning, this paradigm changes, where each student’s device
becomes a decentralized node, storing its data locally via their smartphone.
In our EFL class, we teach our students to use AI applications with a virtual
agent approach, robot tutor, or language learning platform on their respective
devices, smartphones, or laptops, thus enabling the creation of personalized learn-
ing pathways for each undergraduate student. As students engage with EFL lan-
guage practice, the model adapts to their strengths and weaknesses. For example,
if a student excels in vocabulary but struggles with pronunciation, the federated
learning model will of course adapt the exercises to meet their specific needs in
EFL, and vice versa. If an undergraduate student has difficulty with grammar, then
the adjustments to the exercises are done practically. In our EFL scenario, students’
interactions with language applications generate insights locally on their devices,
where federated learning promotes collaborative learning experiences. Insights
gained from each device contribute to collective model refinement. We present fed-
erated learning principles, such as grouping our students who excel in a particular
language skill, so the federated learning model (FLM) will adapt to combine these
18 Artificial Intelligence Using Federated Learning

successful strategies, thereby providing benefits for the entire class and according
to the needs of each group.
AI virtual robotic tutor Lily and Elsa platforms are used in language learning that
utilizes the federated learning model to provide adaptive content recommendations.
As students progress, the model refines its understanding of their preferences and
learning styles. We identify, for example, that when a student consistently engages
in conversational practice, the model will suggest more interactive speaking activi-
ties to improve fluency, and the student can activate the audio feature to practice
and continually refine its language proficiency predictions, ensuring adaptability to
evolving classroom needs.
Therefore, we were glad to see the target and changes in language teaching by
decentralizing the learning process, personalizing teaching, preserving privacy,
encouraging collaboration, and providing real-time feedback. In our analysis, this
fresh method heralds a sea change in language instruction and points the way toward
the development of more effective and flexible classroom settings for English as a
Foreign Language students. By working together to train machine learning models,
academic institutions can protect students’ personal information while also collecting
and sharing expertise in a certain area. Industry 6.0 participants may work together
more effectively and share more information. This promotes collective intelligence
and innovation in various industries (Zeng et al., 2022). The framework can also
make use of FLM to promote the ethical and responsible usage of AI. To promote AI
models that are just transparent and accountable, collaboration is paramount. This
will help to actively prevent any type of unethical behavior, such as prejudice or bias.
This strategy can be used to create Industry 6.0 AI systems that are more trustworthy
and responsible. These transitions in education also will help to make sure that this
technology is used in an ethical way and following social norms.
Integrated human–robot FLM based on collaborative human–robot integration.
One important part of working together is the ability to communicate and analyze
data in real time within the language domain. More collaborative integration between
humans and robots is a logical outcome of users being able to enhance their language
abilities in a complex learning environment that is customized to their needs. It makes
it easier to train machine learning models that can be adjusted to meet the unique
requirements of each student, which in turn allows for the creation of highly custom-
ized and personalized learning. As the focus here is on actual 3D capabilities and
other cutting-edge production technologies, humans and machines may work together
to create one-of-a-kind products (Jia & Liu, 2019; Yakin et al., 2022).

FEDERATED LEARNING MODEL IN ENGLISH CLASS

Indonesia is one of the countries where people who do not speak English must now
have a good understanding of English with the correct communication skills (micro
and macro learning) because English is the language of global communication. One
needs a reliable tool that can automate the training and identification process to
improve their English skills. In the field of English education, the challenge is to be
able to get tailored instruction to improve their English skills with tools that can help
them determine where they are having difficulties. The tool must be accessible, easy
In-Depth Analysis of Artificial Intelligence Practices 19

to use, and embedded in students’ smartphones and accommodate students with var-
ious needs whenever and wherever they are. It would be great if this tool could also
provide instant feedback to users, allowing them to quickly correct their language
errors independently. Recent technological advancements have brought forth a range
of products designed to streamline English language learning through automation.
According to Chen et al. (2019), federated learning is a method whereby small
devices collaborate by sharing computing resources and local data to construct reli-
able models on distributed systems that operate on a large scale. The objective of FL
is to discover answers to several issues that computers aren’t capable of handling,
including those involving users’ private data, computing in real time, and AI robots
embedded in devices (Shakeer & Babu, 2024). A global machine learning model can
be trained using local weights from each user’s data using federated learning, which
enables numerous users to collaborate without revealing any sensitive information
(Weller et al., 2022). This method is employed to achieve a high level of precision in
the global model, as stated by Liu et al. (2020) and Narayan et al. (2016). Yin et al.
(2021) demonstrated that Google Translate’s federated learning (FL) approach in
Florida effectively mitigates privacy risks and reduces data exchange costs while
adhering to GDPR regulations, outperforming traditional centralized machine learn-
ing methods (Lyu et al., 2020). Several studies have compared deep neural networks
(DNNs) with FL (Taik et al., 2024) due to the widespread use of DNN in the field
of machine learning, but there are still not many articles comparing FL with several
other learning techniques, including in the educational realm.
The encouraging findings on DNNs have increased their appeal in various appli-
cations. The integration of DNN into FL is not without problems, especially because
of one obstacle. In their research, Mazzocca et al. (2024) explore potential develop-
ments and future perspectives for federated learning that protect privacy and data. Li
et al. (2021) also compared several FL systems. In addition to providing an overview
of FL, this paper also focuses on its application to image data security, collaborative
AI, and tools in learning EFL. Only a few studies have examined possible barriers to
the use of FL in EFL, even though previous studies have examined the advantages of
FL over classical machine learning in terms of protection, privacy, and convergence
efficiency (Sánchez et al., 2024). To address this gap, this chapter examines existing
research on the subject, categorizes the challenges of FL in foreign language educa-
tion and suggests potential solutions. Despite limited research on the application of
FLM in EFL, this chapter aims to contribute to its exploration.
The AI robotics application can understand users’ speech patterns and provide
them with tailored recommendations by combining machine learning algorithms
with voice recognition technology, among other features. In fields such as education,
business, and customer service, where proper English communication is essential,
the need for such tools is always increasing. Recent advances in deep learning have
significantly increased the effectiveness and precision of systems that handle written
and spoken language. The idea of end-to-end learning is one approach; this requires
teaching a single neural network to translate between unprocessed audio data and
literal text transcription (Hagiwara, 2021; Eslit, 2023). This method has shown
encouraging results in tasks that involve understanding written language, namely
vocabulary, grammar, and spoken recognition, with automatic speech recognition
20 Artificial Intelligence Using Federated Learning

and speech synthesis features. One such idea is the attention mechanism, which
allows the model to pay close attention to important parts of the input sequence. As
a result, results in areas such as speech recognition and translation have improved.
In addition, the capacity to represent long-term dependencies in a series of improv-
ing language skills has been enhanced using robotics-based teaching models, so that
English language processing is increasingly modern and interactive (Wang, 2023).
To improve English language proficiency, robotic learning approaches are now
being used in this study. Therefore, there is an urgent need for intelligent and real-
time help systems to improve English language proficiency. The suggested work
revolves around natural language processing. We present artificial intelligence and
mobile applications with support for multiple users in this article. When training this
model with a federated learning approach on the aggregated database, we also con-
sider data privacy issues. Therefore, this work aims to develop privacy-preserving
mobile and web multi-app training based on blended learning to assist students in
improving their English language proficiency. One of the main reasons for suggest-
ing blended learning-based systems is so that they can incorporate future technolo-
gies that will expand the possibilities of related applications.

FEDERATED LEARNING: A TAXONOMY

Federated learning is a framework for language training that emerged out of the
intersection of machine learning and rising concerns about data privacy and secu-
rity. The necessity to remedy the drawbacks of traditional machine learning models,
particularly when handling complex and large data sets, led to this categorization.
According to Zhang et al. (2021), there are three distinct types of FL solutions.
Designed with privacy and computational efficiency in mind, these types are tailor-
made for dealing with massive data sets. Let’s have a look at the many forms of
federated learning and describe them:

1. Horizontal federated learning. To define HFL, it is a process whereby

numerous entities work together, with each entity having its local data set
that contains identical features. It enables the simultaneous training of a
model on many data sets independently, eliminating the necessity for data
exchange. Although the model incorporates the collective wisdom of all
involved parties, each one keeps full authority over its data (Beltrán et al.,
2023). A common scenario in EFL classrooms involves students from
diverse academic backgrounds employing a variety of mobile applications
to enhance their grammar proficiency. The data, consisting of responses
to similar questions, is scattered among various platforms such as Lily,
Elsa-AI, Google translation, Google sites, and MindMeister© apps. By col-
lecting knowledge from many different students, horizontal FL can improve
the precision of grammatical models.
2. Vertical federated learning (VFL) is defined as a collection of practices
that apply when two or more parties have complementary knowledge about
a shared data set characteristic. This class makes it easier to train models
on data sets that do not share any characteristics (Duan et al., 2022). By
In-Depth Analysis of Artificial Intelligence Practices 21

limiting data sharing to what is essential, parties can protect the privacy
of their data set components. In language learning programs, for example,
users can choose their vocabulary and save it on their device by using the
MindMeister© application. Without revealing users’ specific word choices,
vertical FL can be used to improve language models collaboratively, for
example, by collecting information from various applications, and the plat-
form can propose customized vocabulary exercises.
3. Last, TFL stands for federated learning transfer. Efficiency gains dur-
ing model training through the incorporation of previously trained global
model information into locally targeted models is one possible interpreta-
tion of transfer learning. With the use of a large data set, TFL trains a global
model. Subsequently, local models refine their performance using their data
sets. Maximizing learning efficiency without compromising data security
is possible with this strategy (Chen et al., 2023). Sentiment analysis engines
are trained to correctly detect the tone of English text by initially exposing
them to data in that language. This is because these engines are designed
to improve learning performance on multiple distributions by transferring
knowledge from one distribution to another. To train a model on English
data, one can use FL transfer. To optimize learning across different lan-
guage distributions, knowledge learned from English data is transferred and
adapted to improve sentiment analysis performance on English knowledge.

FL is related to NLP theory and the theory of human language acquisition, namely
that humans convey a large amount of information through every word and syllable
they pronounce when speaking. Natural language communication is enhanced and
influenced by various factors, such as the subject matter and the sophistication of the
vocabulary used. To enhance language acquisition efficiency and reduce the emotional
strain on teachers and students, it is crucial to explore the potential of AI. Students can
learn independently, even without a teacher, because AI is set to be available 24/7. The
traditional concept of remembering and evaluating the structure, language, meaning,
and unique tone of each sentence is an impractical idea (Liu & Lu, 2023).
As can be seen, one branch of artificial intelligence offers machines that can read,
understand, predict, and extract meaning from human spoken language, and this
intelligence is of great benefit to EFL students (Celik, 2023). One common applica-
tion of this field is sentiment analysis, which involves identifying and classifying the
emotional tone of text, such as positive, negative, or neutral. Information is provided
regarding factors that influence student preferences. With the growing popularity of
sentiment analysis on the web, individuals can learn from multiple perspectives and
stories. Through social media, more and more people are sharing their opinions with
strangers by tweeting about a variety of subjects (such as politics, consumer prefer-
ences, vacations, and other topics where the app excels).
Expanding to support the collaboration of FLM for EFL, data translation is known
as machine translation (Benbada & Benaouda, 2023), which uses machines to trans-
late different languages as students become more adept at translating using technol-
ogy. Google Translate is the most well-known example of an application that uses
statistical machine learning (SML) language processing for EFL, and this engine
22 Artificial Intelligence Using Federated Learning

was very popular for students before the advent of AI applications for language prac-
tice such as Lily Tutor and Elsa AI. The basic premise is to collect a set of possible
texts, looking for similar texts in opposing languages, with the possibility of them
being translated into another language. Computer speech recognition, or autono-
mous speech recognition (ASR), is another name for speech recognition such as Elsa
AI. It involves carrying out verbatim transcription of a speech input stream through a
program programmed into a computer system. With proper pre-processing, some of
these uses of natural language processing can become plain text that can be parsed in
a way that any computer can parse. This includes describing a sequence of numeric
words as input. There needs to be a numerical representation that conveys multiple
verbal meanings while maintaining semantic integrity for each syllable. There is a
contemporary paradigm in NLP where the main method is to achieve this.

METHOD AND PARTICIPANTS

The research methodology adopted in this research is an experimental method,
specifically using a pre-experimental design (Marsden & Torgerson, 2012). This
categorization is due to the initial nature of the design, which does not have the
characteristics of a true experiment due to the presence of external variables that
can influence the dependent variable. Consequently, the experimental results, which
represent the dependent variable, are not exclusively governed by the independent
variables. This limitation arises due to the absence of control variables and non-
random sample selection.
The specific research design used is the one-group pretest-posttest design. In
this design, a pretest is given before implementing the treatment, which aims to
provide a baseline measure. This approach increases the precision of the assess-
ment of treatment effects by facilitating comparison with pretreatment conditions.
The design is characterized by giving a pretest (initial test) and posttest (final test)
sequentially to a single group. The research procedure steps use a one-group pretest
and posttest design, which includes measuring the dependent variable (pretest), pro-
viding treatment (X) for six weeks involving students by implementing a federated
learning approach using the Tutor Lily and Elsa AI applications, as well as subse-
quent measurement of the dependent variable after treatment, as shown in Table 2.1.
To operationalize the one-group pretest-posttest design, a pretest (O1) was given
to students in semester 1 of the 2023 academic year to ensure their initial knowl-
edge before implementing the treatment. Next, treatment was carried out in the form
of implementing federated learning in AI-integrated classes to improve learning
outcomes on the concept of English language skills. The difference in initial and

TABLE 2.1
Pre-Experimental Design
01 X 02
Pretest Treatment Post Test
In-Depth Analysis of Artificial Intelligence Practices 23

FIGURE 2.1 FLM approach.

final test scores (O1 and O2) is assumed to represent the effect of treatment (X).
This design is in line with the research objective, assessing improvements in student
learning outcomes after implementing the AI-based FL model.
To describe the research context, the population is undergraduate students at Al
Asyariah Mandar University, totaling 219 students. The sample selection involved 13
study programs: Indonesian language education, mathematics education, Pancasila
and citizenship education, public health, sharia economics, Islamic management,
agrotechnology, agribusiness, animal husbandry, communication science, gov-
ernment science, informatics engineering, and informatics systems. The samples
exposed to AI were the communication study program, animal husbandry study
program, public health study program, and agrotechnology study program for each
learning model, so a total of four experimental classes were obtained with a sample
of 74 undergraduate students. The research took place in the first (odd) semester of
the 2023 academic year, and data collection was carried out six times in the experi-
mental class, as shown in Figure 2.1.

ADOPTING FL AI-DRIVEN PEDAGOGY

Among a sample of first-semester undergraduate students combining two AI tutors,
our main aim was to identify the impact of robotics learning activities with a
federated learning approach on English language skills. Participants were given
treatment for eight meetings. The population of this study was 219 undergraduate
students with beginner-level English skills. The sample consisted of 74 students
with an age range of 18–21 years, 11 males and 63 females. Participants attended
eight English lectures in class with AI interaction by the learning syntax, as shown
in Table 2.2. The design of this research was quantitative with a pre-experiment or
did not use a control class.
This English language teaching model provides an innovative and technology-
infused language teaching approach, utilizing blended learning and AI robot tutors
with a focus on English language skills. Undergraduate students engage in inter-
active and personalized activities, thereby fostering a comprehensive and effective
language learning experience that can enhance learning outcomes.
24 Artificial Intelligence Using Federated Learning

TABLE 2.2
Teaching Model Incorporating FL
Step 1. In the first stage, undergraduate students are introduced to the concepts of robotics-based
Warmup learning and federated learning, which involve conversations with AI robot tutors to develop
conversation skills. Undergraduate students are asked to engage in interactive dialogue with
an AI tutor, with a focus on pronunciation, fluency, and contextual communication. Start by
downloading the Robot Tutor application on the student’s smartphone.
Step 2. Undergraduate students are asked to form groups and discuss with their group friends the
Federated conversation theme, the weather. They discussed determining weather keywords by
combining robot tutor Lily and Mind Mapping. At this stage, undergraduate students are
asked to access Tutor Lily, which helps them discover vocabulary related to weather. After
that, they organized their group’s vocabulary according to the weather conditions that Tutor
Lily shared. Undergraduate students will practice pronunciation of vocabulary for those who
are not yet able to pronounce the vocabulary, practice accents, and practice correct writing.
Step 3. Under Tutor Lily’s guidance, students collaborate with their peers to create practice scenarios
Composition for writing dialogues. These exercises are designed to improve grammar, coherence, and
vocabulary, with AI providing feedback. In this stage, undergraduate students access L2
using Tutor Lily’s translation features and Google Translation. To perfect the dialogue,
lecturers guide undergraduate students and facilitate with Grammarly. This activity trains
collaborative grammar that utilizes blended learning, ensuring collective improvement in
grammatical accuracy. In addition, at this stage, undergraduate students structurally enrich
their vocabulary through AI-guided exercises and contextual usage exercises.
Step 4. Practice stage: after the dialogue is complete, it continues with personalized reading with
Practice AI with the aim of improving reading comprehension through a personalized and
AI-guided reading experience. A dialogue script about the presented material (weather)
makes up the reading material that the AI has curated. Together with their group mates,
they practice with each other to sharpen their listening skills with AI-generated audio
content. In addition to Tutor Lily, undergraduate students access Elsa-AI. This stage of
AI integration aims to train pronunciation, practice listening, and practice pronunciation
with the help of AI and assess understanding.
Step 5. In the performance stage, undergraduate students and their group friends display their
Performance English language skills by recording a video presentation and uploading it to YouTube.

RESEARCH INSTRUMENTS
This research uses a combination of test and non-test instruments to collect data for
comprehensive analysis. The testing component consists of a multiple-choice ques-
tionnaire consisting of 25 questions given at the initial stage. This test had a dual
purpose, assessing the initial proficiency of both experimental groups and establish-
ing the homogeneity of their abilities. This careful assessment aims to ensure that, at
the start of the study, the two groups do not show statistically significant differences
in their capacities, as shown in Figure 2.2.
To assess whether the experimental groups had made any improvements in their
learning outcomes, they were both given an identical multiple-choice questionnaire.
To fully grasp the efficacy of the interventions used in the study, this cyclical testing
method is employed.
In-Depth Analysis of Artificial Intelligence Practices 25

Process
Input 1. Treatment AI Output
1. Pretest Robotic on EFL 1. Posttest
2. Observation 2. English skills 2. Data Analysis
3. FLM-based learning

Dissemination
Outcome YouTube

FIGURE 2.2 Research procedure.

A psychomotor evaluation, which is not a test but is administered at the same time
as the test instrument, is also conducted. The goal of this all-encompassing strategy
is to gain a thorough knowledge of the many facets of student learning outcomes by
integrating test and non-test instruments. After carefully administering the instru-
ments and collecting the data, as well as making sure that the instruments were valid
and appropriate for the research purpose, a series of advanced data analyses were car-
ried out. To conduct a thorough and rigorous review of the treatments used and their
effect on the experimental group’s learning outcomes, this study seeks to acquire new
insights into the complexities of the research findings shown in Figure 2.3.

DATA ANALYSIS
After a comprehensive verification process and extensive tests for validity and com-
pliance with set standards, the next crucial step is to quantify the research results.
Two sophisticated techniques used in the analysis process are quantitative descrip-
tive analysis and inferential statistics, which are both made possible by SPSS version
26. Thoroughly analyzing and making sense of the numerical data produced by the
instrument is essential for quantitative descriptive analysis. Using statistics like the
mean, median, and standard deviation, one can get a good idea of the dataset’s cen-
tral tendency and dispersion and then conclude the set’s key properties. At the same
time, relevant inferences are drawn using inferential statistics and dispersion in the
collected data. Simultaneously, inferential statistics are used to draw meaningful
conclusions and infer patterns in datasets. The use of SPSS version 26, powerful
statistical software, allows the application of various statistical tests.

RESULTS AND DISCUSSION

Using descriptive statistics, we were able to better understand the distribution and
central tendencies of performance indicators in the data set that included pre-and
posttest scores from 74 participants. Participants’ baseline performance on the
26 Artificial Intelligence Using Federated Learning
FIGURE 2.3 Extracting performance in YouTube.
In-Depth Analysis of Artificial Intelligence Practices 27

TABLE 2.3
Pre- and Post-Test Results
N Minimum Maximum Mean Std. Deviation
Pretest 74 30.00 70.00 54.3243 8.96433
Posttest 74 60.00 95.00 79.6622 9.52038
Valid N (listwise) 74

TABLE 2.4
Chi-Square Test
Pretest Posttest
Chi-square 58.811a 41.459b
df 8 7
Asymp. sig. .000 .000
a 0 cells (0.0%) have expected frequencies less than 5. The minimum expected cell frequency is 8.2.
b 0 cells (0.0%) have expected frequencies less than 5. The minimum expected cell frequency is 9.3.

pretest showed a substantial amount of variability, with scores ranging from 30 to 70

on the scale in Table 2.3.
Table 2.3 shows the mean score was 54.3243, and the standard deviation was
8.96433. Afterward, on the posttest, the scores varied from 60.00 to 95.00, showing
an increasing tendency. There was a 9.52038 standard deviation and a significant
increase in the mean posttest score to 79.6622. Both evaluations use the same valid
N (listwise) value of 74, which shows that the data set is comprehensive.
The relationship between pretest and posttest scores was assessed using statisti-
cal tests, specifically the chi-square test. The results in Table 2.4 show that there are
statistically significant consequences. With 8 degrees of freedom, the pretest chi-
square value is 58.811, resulting in a p-value of 0.000, which is considered highly
significant. Apart from that, the posttest shows a p-value of 0.000, 7 degrees of
freedom, and a chi-square value of 41.459. It is worth mentioning that in both cases,
zero cells (0.0%) meet the minimum predicted cell frequency requirement, with an
expected frequency below five. For the pretest, the lowest predicted cell frequency
was recorded at 8.2, and for the posttest, it was 9.3.
Table 2.5 shows the mean difference of 54.32432 from the test results is statisti-
cally significant, according to the pretest, which has a t-statistic of 52.131 with 73
degrees of freedom and a p-value of 0.000. The 95% confidence interval of the dif-
ference (52.2475, 56.4012) further supports these findings. The posttest also shows
a t-statistic of 71.980 with 73 degrees of freedom and a significant p-value of 0.000,
which means the mean difference of 79.66216 from the test value of 0 is statistically
significant. The 95% confidence interval accompanying the difference (77.4565,
81.8679) further indicates the dependence of this effect.
28 Artificial Intelligence Using Federated Learning

TABLE 2.5
One Sample Test
Test Value = 0
95% Confidence
Interval of the
Mean Difference
t df Sig. (2-Tailed) Difference Lower Upper
Pretest 52.131 73 .000 54.32432 52.2475 56.4012
Posttest 71.980 73 .000 79.66216 77.4565 81.8679

TABLE 2.6
ANOVA
Sum of Squares df Mean Square F Sig.
Between Groups 1707.894 7 243.985 3.872 .001
Within Groups 4158.322 66 63.005
Total 5866.216 73

TABLE 2.7
Paired Sample Correlations
N Correlation Sig.
Pair 1 Pretest and posttest 74 .403 .000

To determine whether the pretest scores varied significantly between catego-

ries, we used analysis of variance (ANOVA), as shown in Table 2.6. We found that
the between-group variability, which represents differences among the groups,
accounted for 1707.894 sums of squares with 7 degrees of freedom, giving us a
mean square of 243.985 when we examined the sources of variation. The estimated
F-statistic of 3.872 and its related p-value of.001 among the groups demonstrated a
statistically significant difference. Alternatively, there was a mean square of 63.005
and a sum of squares of 4158.322 for the within-group variability, which reflects
individual differences within each group. There were 66 degrees of freedom in this
analysis. With a total of 73 degrees of freedom, the sum of squares for the pretest
scores was 5866.216. The statistical significance of the between-group F-ratio indi-
cates that there are obvious group-wise variations in the pre-test results.
Table 2.7 shows the results of a paired sample correlation analysis conducted
on 74 individuals to examine the relationship between pretest and posttest scores.
Calculation of the correlation coefficient (r) between the pretest and posttest scores
shows that there is a quite positive relationship, with a value of 0.403. The statisti-
cal significance of the correlation, with a p-value of 0.000, indicates that it is very
unlikely that the observed relationship occurred by chance, as shown in Figure 2.4.
In-Depth Analysis of Artificial Intelligence Practices 29
FIGURE 2.4 The histogram pretest and posttest.
30 Artificial Intelligence Using Federated Learning

Table 2.8 shows the survey data set includes descriptive statistics for variables related
to EFL classroom pedagogical considerations, AI practices, robotic tutor involvement,
blended learning implementation, AI practices, and blended learning integration, as
well as general pedagogical considerations. Primary statistical measures were applied
to participants’ responses after each variable was evaluated using various statements.
For AI practices, participants demonstrated a strong understanding of the subject
(mean = 4.5270) and actively engaged with AI-enabled technologies (mean = 4.4189).

TABLE 2.8
Questionnaire Results
Descriptive Statistics
Std.
N Minimum Maximum Mean Deviation
AI Practices
1. I understand the practice of artificial intelligence 74 3.00 5.00 4.5270 .68692
(AI) in education, and I am happy
2. I use AI-powered tools or applications in English 74 2.00 5.00 4.4189 .77648
as a Foreign Language (EFL) learning very well.
3. From my perspective, AI-driven tools are a promising 74 2.00 5.00 4.1892 .87077
avenue for optimizing language acquisition.
4. AI Tutor Lily and Elsa’s technology or practice 74 1.00 5.00 4.1486 .85500
integrated into EFL lessons motivates interest in
learning English compared to before without using AI.
5. I believe that AI-based practices can contribute to 74 2.00 5.00 4.3378 .83218
individualized learning experiences in EFL
classrooms.
Robot Tutor
Using a robot tutor in EFL class is very interesting. 74 2.00 5.00 4.3243 .74223
I have found that incorporating a robot tutor into my 74 2.00 5.00 4.3649 .82053
English learning has been very beneficial.
I am more comfortable learning when lecturers use 74 2.00 5.00 4.2027 .84367
Robot Tutor Lily and ELAS compared to traditional
teaching methods.
My language skills can be improved effectively with 74 1.00 5.00 4.1216 .92088
the help of AI.
Implementation of federated learning
I enjoy learning English with the concept of federated 74 1.00 5.00 4.1351 .95551
learning (FL) in the context of language education.
Learning federation in EFL class can help my deep 74 2.00 5.00 4.2297 .86875
learning of English.
I feel the potential benefits when lecturers use 74 1.00 5.00 4.3919 .82458
federated learning for language teaching.
I am excited about the federated learning model in an 74 1.00 5.00 4.2162 .86437
EFL environment.
I believe that federated learning can address privacy 74 2.00 5.00 4.2973 .83960
issues in language education.
In-Depth Analysis of Artificial Intelligence Practices 31

TABLE 2.8 (Continued)

Questionnaire Results
Descriptive Statistics
Std.
N Minimum Maximum Mean Deviation
Integration of AI practices and federated learning
Integrating AI practices and federated learning in EFL 74 1.00 5.00 4.2162 .89550
teaching helps me to analyze language, from
vocabulary to designing practical sentences and
dialogues.
I felt the synergy between AI practices and federated 74 2.00 5.00 4.1622 .92198
learning as being able to optimize language learning
outcomes.
I like to learn English by using the AI and FL 74 1.00 5.00 4.0811 1.03045
approaches.
Pedagogical considerations:
I am sure that the integration of robot tutoring, artificial 74 2.00 5.00 4.1486 .90179
intelligence (AI) practices, and federated learning can
achieve maximum learning outcomes.
I believe in the impact of AI practices, especially robot 74 2.00 5.00 4.1757 .88144
tutoring, on encouraging student engagement and
participation in EFL lessons.
In my opinion, educators should adapt their teaching 74 1.00 5.00 4.1486 .96063
strategies to accommodate the incorporation of AI
and federated learning practices in EFL teaching.
I recommend that educators incorporate AI practices 74 2.00 5.00 4.1757 .86576
and federated learning into teaching strategies for
EFL and other courses.
Valid N (listwise) 74

DISCUSSION
As stated in Table 2.3, the data of descriptive statistics provide a detailed picture of
the distribution of participants’ performance, which can be used as a starting point for
inferential studies and to guide possible educational initiatives. Based on Table 2.4,
the results of the chi-square test showed that the association test was strong, indi-
cating a correlation between pre- and posttest scores. This can help us understand
educational interventions or experimental conditions better. Using a test value of 0,
a one-sample test was performed on the pretest and posttest data sets to see whether
the mean differences in each group were statistically significant. The results showed
that there was a statistically significant change between the mean scores before and
after the test, which explains how well the intervention of the FL teaching method
worked. A substantial drop in performance between the two sets of assessments
suggests the need for additional research into the intervention’s efficacy or the effec-
tiveness of the teaching method. Consequently, the coefficient value, which shows
32 Artificial Intelligence Using Federated Learning

the strength of the link, reveals that participants with higher pretest scores also tend
to receive higher posttest scores. This study sheds light on people’s performance on
both exams and can provide light on the kinds of educational interventions or experi-
mental conditions that may have produced the observed correlations.
Based on Table 8, the mean scores of students’ responses from the survey ranged
from 4.1216 to 4.3919. This indicated that undergraduate students were generally
positive in terms of their experiences with Tutor Robot and the application of fed-
erated learning in the EFL classroom. Positive feedback was also received by AI
practice integration and federated learning, highlighting the real capacity of AI
practice integration and federated learning model to improve language learning out-
comes such as Bowser’s added-on (Farahrani et al., 2023), which trains algorithms
using distributed data sets stored on edge devices or servers. This proposed method
eliminates the need to communicate data, thereby bridging the privacy utility gap
introducing a new personalized clustered FL (PCFL) method that trains models by
leveraging shared features in the personal data and by capturing unique features
in each client’s data. Although the proposed methodology successfully overcomes
privacy concerns, experimental findings show that it hardly loses performance when
compared with centralized model training methods.
Figure 2.2 shows the teaching model taxonomy by integrating the FL concept,
which is based on the FL classification, namely horizontal federated learning, vertical
federated learning, and federated learning transfer, and the AI application domain.
Segmentation of the language data obtained is distributed to test EFL English skills,
and the personalized approach is part of the teaching model that is tested in the
classroom. Some of the main obstacles include quota costs, network constraints, the
diversity of AI systems in use, and concerns about privacy and security. Analysis of
survey paper topics and their correlation with language learning outcomes suggest
potential improvements in FL implementation through categorization. Moreover, a
majority of participants advocated for the integration of AI into English as a Foreign
Language classrooms, echoing the positive findings on student engagement reported
by Braines et al. (2017).
In addition to the challenges faced, the study revealed that undergraduate students
perceived FLM as a valuable tool for enhancing their EFL language skills, consistent
with previous research (Chhikara et al., 2020; Purgina, 2019). Their studies showed
that the use of various artificial intelligence applications can support language train-
ing speed, efficiency, generalization ability, and suitability for various domains, such
as the Internet of Things (IoT), Google Translation collaboration, storing data on
vocabulary, grammar, sentences obtained on smartphone devices, notes, and secu-
rity data for each undergraduate student. Integrating FL in class has its own chal-
lenges. In this study, this is the first time this has been done and required repeated
explanations so that undergraduate students could understand each learning syntax
structure and adapt well. Priorities, including optimal internet networks and differ-
ent types of smartphones, certainly do not hinder FL acceleration in class.
When designing learning scenarios with the integration of Tutor Lily and Elsa AI
(customized models depending on the learning environment), personalization is an
additional component to consider. Barriers to implementing an FL-based learning
system are undergraduate students’ high internet costs and the need for data storage
In-Depth Analysis of Artificial Intelligence Practices 33

space on their respective smartphones. Because of the large number of devices that
may form a combined network, data transfer over the network is much slower than
computing performed locally. As a result, effective networking has been the subject
of much research in Indonesia and the Asia Pacific in general. The diversity of under-
lying systems also means that devices in a combined network may have different
communication capabilities. Different devices may also have varying amounts of
storage and processing power, with some downloading quickly and others being slow.
In many circumstances, network limitations do not mean that only some devices can
take part in an iteration. Therefore, the application of learning techniques with the
FL approach can overcome and find solutions to these obstacles.
The adoption of combined learning methods, similar to those used in education,
is still uncommon in fields like engineering, architecture, IoT, and health, where the
term “FL” is predominantly used. Several elements contribute to improving learning
outcomes with FL techniques; apart from technology and teaching models, condi-
tions and consideration of students’ learning needs are also considered. The time
required for blended learning to come together considers not only the processing
power of language acquisition but also the time required to adapt to the technique.
Therefore, to implement and optimize the FL approach, it is important to consider
the quality of the internet, student motivation, and learning experience so that this
learning model works well and can be applied to other courses. To optimize the
combined deep learning model for AI applications, it can be carried out adaptively,
considering the availability of devices and their readiness to participate in the learn-
ing process.

CONCLUSION
This research incorporates a federated learning approach in language learning by
using three FL characters in learning activities to see how federated mobile learning
can help overcome the problem of low English language skills, interest, and learning
motivation, especially those related to the anywhere and anytime learning paradigm.
As an innovative decentralized learning solution, the FL approach was tested in
language classes to overcome the problems of decentralized EFL language learning.
We introduce the concept of federated learning and the idea of blended learning and
its main characteristics in a way that is easy for everyone to understand. The results
of our study outline the steps for implementing blended learning in language classes
and the possibility that it can be adapted for other courses. Moreover, this research
shows that blended learning can be used with simulations that have been run for
eight meetings, and our results show that blended learning techniques can improve
learning outcomes, motivation, and interest in learning English. Additional studies
are needed to address the many unresolved concerns and barriers.
The application of FL in the field of language can be done with the concept of a
collaborative human–robot integration approach. Collaboration occurs by emphasiz-
ing real-time features in data sharing and analysis. A more collaborative integration
of humans and robots is of course supported by a more flexible learning environ-
ment, allowing users to improve their language skills according to their needs in a
sophisticated learning environment. Strategies based on collaborative human–robot
34 Artificial Intelligence Using Federated Learning

integration can be used to implement FL in the language domain. Real-time data

exchange and analysis is a key component of collaboration. Users can improve their
language skills in a complex learning environment tailored to their needs, which of
course supports more collaborative integration between humans and robots.

REFERENCES
Al Yakin, A., Obaid, A. J., Muthmainnah, R. S. A. M., Khalaf, H. A., & Al-Barzinji, S. M.
(2022). Bringing technology into the classroom amid Covid 19, challenge and opportu-
nity. Journal of Positive School Psychology, 6(2), 1043–1052.
Angurala, M., & Khullar, V. (2023). Federated learning-based privacy preserved English
accent training ecosystem for people with Indian language accent. Entertainment
Computing, 46, 100572.
Beltrán, E. T. M., Pérez, M. Q., Sánchez, P. M. S., Bernal, S. L., Bovet, G., Pérez, M. G.,. . .&
Celdrán, A. H. (2023). Decentralized federated learning: Fundamentals, state of the
art, frameworks, trends, and challenges. IEEE Communications Surveys & Tutorials,
25(4), 2983–3013. http://dx.doi.org/10.1109/COMST.2023.3315746
Benbada, M. L., & Benaouda, N. (2023). Investigation of the Role of Artificial Intelligence
in Developing Machine Translation Quality. Case Study: Reverso Context and Google
Translate Translations of Expressive and Descriptive Texts. Language Combination:
Arabic-English/English-Arabic (Doctoral dissertation, Faculty of Letters and Languages-
Department of English).
Braines, D., O’Leary, N., Thomas, A., Harborne, D., Preece, A. D., & Webberley, W. M.
(2017). Conversational homes: A uniform natural language approach for collaboration
among humans and devices. International Journal on Advances in Intelligent Systems,
10(3/4), 223–237.
Celik, I. (2023). Towards intelligent-TPACK: An empirical study on teachers’ professional
knowledge to ethically integrate artificial intelligence (AI)-based tools into education.
Computers in Human Behavior, 138, 107468.
Chen, D., Jiang, X., Zhong, H., & Cui, J. (2023). Building trusted federated learning: Key
technologies and challenges. Journal of Sensor and Actuator Networks, 12(1), 13.
Chen, M., Suresh, A. T., Mathews, R., Wong, A., Allauzen, C., Beaufays, F., & Riley, M. (2019).
Federated learning of n-gram language models. arXiv preprint arXiv:1910.03432.
Chhikara, P., Singh, P., Tekchandani, R., Kumar, N., & Guizani, M. (2020). Federated learn-
ing meets human emotions: A decentralized framework for human–computer interac-
tion for IoT applications. IEEE Internet of Things Journal, 8(8), 6949–6962.
Chourasia, S., Tyagi, A., Pandey, S. M., Walia, R. S., & Murtaza, Q. (2022). Sustainability
of industry 6.0 in global perspective: Benefits and challenges. Mapan, 37(2), 443–452.
de Moraes Rossetto, A. G., Martins, T. C., Silva, L. A., Leithardt, D. R., Bermejo-Gil, B.
M., & Leithardt, V. R. (2023, November). An analysis of the use of augmented reality
and virtual reality as educational resources. Computer Applications in Engineering
Education, 31(6), 1761–1775. https://doi.org/10.1002/cae.22671
Driss, M. B., Sabir, E., Elbiaze, H., & Saad, W. (2023). Federated learning for 6G: Paradigms,
taxonomy, recent advances and insights. arXiv preprint arXiv:2312.04688.
Duan, Q., Hu, S., Deng, R., & Lu, Z. (2022). Combined federated and split learning in edge
computing for ubiquitous intelligence in internet of things: State-of-the-art and future
directions. Sensors, 22(16), 5983.
Eslit, E. R. (2023). Elevating language acquisition through deep learning and meaningful
pedagogy in an AI-evolving educational landscape. Preprints. https://doi.org/10.20944/
preprints202309.0658.v1
In-Depth Analysis of Artificial Intelligence Practices 35

Farahani, B., Tabibian, S., & Ebrahimi, H. (2023). Towards a personalized clustered federated
learning: A speech recognition case study. IEEE Internet of Things Journal, 10(21),
18553–18562. doi: 10.1109/JIOT.2023.3292797
Hagiwara, M. (2021). Real-World Natural Language Processing: Practical Applications with
Deep Learning. Simon and Schuster.
Jia, T., & Liu, Y. (2019, October). Words in kitchen: An instance of leveraging virtual reality
technology to learn vocabulary. In 2019 IEEE International Symposium on Mixed and
Augmented Reality Adjunct (ISMAR-Adjunct) (pp. 150–155). IEEE.
Khalid, U., Iqbal, H., Vahidian, S., Hua, J., & Chen, C. (2023). CEFHRI: A communication
efficient federated learning framework for recognizing industrial human–robot interac-
tion. arXiv preprint arXiv:2308.14965.
Khang, A., Muthmainnah, M., Seraj, P. M. I., Al Yakin, A., & Obaid, A. J. (2023a). AI-aided
teaching model in education 5.0. In Handbook of Research on AI-Based Technologies
and Applications in the Era of the Metaverse (pp. 83–104). IGI Global.
Khang, A., Shah, V., & Rani, S. (Eds.). (2023b). Handbook of Research on AI-Based
Technologies and Applications in the Era of the Metaverse. IGI Global.
Klimova, B., Pikhart, M., Polakova, P., Cerna, M., Yayilgan, S. Y., & Shaikh, S. (2023).
A systematic review on the use of emerging technologies in teaching English as an
applied language at the university level. Systems, 11(1), 42.
Li, Q., Wen, Z., Wu, Z., Hu, S., Wang, N., Li, Y.,. . .& He, B. (2021). A survey on feder-
ated learning systems: Vision, hype and reality for data privacy and protection. IEEE
Transactions on Knowledge and Data Engineering, 35(4), 3347–3366. doi: 10.1109/
TKDE.2021.3124599
Lindblom, J., Alenljung, B., & Billing, E. (2020). Evaluating the user experience of human–
robot interaction. In Human–robot interaction: Evaluation Methods and Their
Standardization (pp. 231–256). Springer.
Liu, B., & Lu, Z. (2023). Design of spoken English teaching based on artificial intelligence
educational robots and wireless network technology. EAI Endorsed Transactions on
Scalable Information Systems, 10(4), e12–e12.
Liu, Y., Yuan, X., Xiong, Z., Kang, J., Wang, X., & Niyato, D. (2020). Federated learning for 6G
communications: Challenges, methods, and future directions. China Communications,
17(9), 105–118.
Lyu, L., Xu, X., Wang, Q., & Yu, H. (2020). Collaborative fairness in federated learning.
Federated learning: Privacy and Incentive, 189–204.
Marsden, E., & Torgerson, C. J. (2012). Single group, pre-and post-test research designs:
Some methodological concerns. Oxford Review of Education, 38(5), 583–616.
Mazzocca, C., Romandini, N., Montanari, R., & Bellavista, P. (2024). Enabling federated
learning at the edge through the IOTA tangle. Future Generation Computer Systems,
152, 17–29.
Muthmainnah, Obaid, A. J., Al Yakin, A., & Brayyich, M. (2023, June). Enhancing computa-
tional thinking based on virtual robot of artificial intelligence modeling in the English
language classroom. In International Conference on Data Analytics & Management
(pp. 1–11). Springer Nature.
Narayan, R., Chakraverty, S., & Singh, V. P. (2016). Quantum neural network based machine
translator for English to Hindi. Applied Soft Computing, 38, 1060–1075.
Papadopoulos, G. T., Antona, M., & Stephanidis, C. (2021). Towards open and expandable
cognitive AI architectures for large-scale multi-agent human-robot collaborative learn-
ing. IEEE Access, 9, 73890–73909.
Purgina, M. (2019). Mobile Technology for Gamification of Natural Language Grammar
Acquisition (Unpublished doctoral dissertation, The University of Aizu). https://uaizu.
repo.nii.ac.jp.
36 Artificial Intelligence Using Federated Learning

Sadiku, M. N., Musa, S. M., & Chukwu, U. C. (2022). Artificial Intelligence in Education.
iUniverse.
Sánchez, P. M. S., Celdrán, A. H., Xie, N., Bovet, G., Pérez, G. M., & Stiller, B. (2024).
Federated trust: A solution for trustworthy federated learning. Future Generation
Computer Systems, 152, 83–98.
Shakeer, S. M., & Babu, M. R. (2024). A study of federated learning with internet of things
for data privacy and security using privacy preserving techniques. Recent Patents on
Engineering, 18(1), 1–17.
Taik, A., Abouaomar, A., & Cherkaoui, S. (2024). Green federated learning-based models
and protocols. In Green Machine Learning Protocols for Future Communication
Networks (pp. 63–102). CRC Press.
Wang, T. (2023, January). Innovative strategies for the development of international Chinese
language education based on deep learning models. Applied Mathematics and
Nonlinear Sciences, 9(1), 1–5. doi: 10.2478/amns.2023.2.00138
Weller, O., Marone, M., Braverman, V., Lawrie, D., & Van Durme, B. (2022). Pretrained
models for multilingual federated learning. arXiv preprint arXiv:2206.02291. https://
doi.org/10.48550/arXiv.2206.02291
Wu, D., Yang, Z., Zhang, P., Wang, R., Yang, B., & Ma, X. (2023). Virtual-reality inter-
promotion technology for metaverse: A survey. IEEE Internet of Things Journal,
10(18), 15788–15809. doi: 10.1109/JIOT.2023.3265848
Yin, X., Zhu, Y., & Hu, J. (2021). A comprehensive survey of privacy-preserving federated
learning: A taxonomy, review, and future directions. ACM Computing Surveys (CSUR),
54(6), 1–36.
Zeng, R., Zeng, C., Wang, X., Li, B., & Chu, X. (2022). Incentive mechanisms in federated
learning and a game-theoretical approach. IEEE Network, 36(6), 229–235.
Zhang, C., Xie, Y., Bai, H., Yu, B., Li, W., & Gao, Y. (2021). A survey on federated learning.
Knowledge-Based Systems, 216, 106775.
 3 Enabling Federated
Learning in the Classroom
Sociotechnical Ecosystem
on Artificial Intelligence
Integration in Educational
Practices
Ahmad Al Yakin, Arkas Viddy, Idi Warsah,
Ali Said Al Matari, Luís Cardoso,
Ahmed A. Elngar, Ahmad J. Obaid,
and Muthmainnah

INTRODUCTION
The introduction of AI technology can affect teacher–student relationships.
Understanding the sociological aspects of this relationship is critical to ensuring that
implementing blended learning enhances, rather than hinders, the educational expe-
rience (Muthmainnah et al., 2023a). Teachers play an important role in mediating the
impact of AI on the classroom environment and fostering a positive sociotechnical
ecosystem. This chapter explores the integration of blended learning into educational
practice, focusing on sociological aspects to see the impact of its implementation in
the classroom. Federated learning, a decentralized machine learning approach, has
great potential to improve educational outcomes. However, its successful implemen-
tation requires a different understanding of the social dynamics in the educational
environment. To investigate the sociological implications of the introduction of arti-
ficial intelligence (AI) through blended learning in the classroom, examining the
potential benefits and challenges from a sociotechnical perspective with the basic
assumption that artificial intelligence has become an integral part of various sectors
and their sectors. Its application in education is receiving increasing attention.
Blended learning, a decentralized machine learning paradigm, offers a unique
approach to incorporating AI in the educational landscape that focuses on the socio-
logical dimensions that enable blended learning in the classroom, highlighting the
complex interactions between technology and social structures Ng et al. (2023).
To understand the implications of AI integration, it is important to take a socio-
logical perspective. Sociological theories, such as symbolic interactionism, social
DOI: 10.1201/9781003482000-3 37
38 Artificial Intelligence Using Federated Learning

constructivism, and sociology of education, provide valuable frameworks for ana-

lyzing the dynamics between students, teachers, and AI technologies in learning
environments (Pedro et al., 2019). The benefits of federated learning in education
address concerns regarding data privacy and security by allowing model training on
decentralized devices. This approach empowers educational institutions to leverage
the benefits of AI without compromising sensitive student information (Rajbongshi
et al., 2022). Additionally, blended learning facilitates a personalized learning expe-
rience by adapting to individual student needs and improving overall academic
performance. Despite its potential benefits, the integration of blended learning
in classrooms poses challenges with sociological implications. Issues such as the
digital divide and ethical considerations surrounding the application of AI require
careful consideration. A sociological perspective helps address these challenges by
examining power dynamics, social stratification, and their impact on diverse student
populations Donnelly (2017). Rapid progress on the Internet of Things (IoT) has
been observed in recent years. This network enables extensive sensing and comput-
ing capabilities to connect various objects to the Internet (Čolaković and Hadžialić,
2018). An approach to developing intelligent and privacy-preserving IoT systems
called federated learning has recently been proposed.
The recently proposed federated learning (FL) technique has attracted the interest
of many scientists who want to learn more about its possibilities and uses (Zhang et al.,
2022). The main goal of FL is to answer the following question: Is it possible to train
a model locally, without transferring data to a central repository? Traditional machine
learning algorithms do not always provide the desired results when working in an
FL framework that emphasizes labor. Another benefit of FL over more conventional
machine learning approaches is that it allows the underlying algorithm to learn from its
errors (Ouyang et al., 2021). FL has been used in various fields and industries, includ-
ing healthcare, the Internet of Things, transportation, and defense, and this chapter
will investigate symbolic interaction on mobile applications used in education.
We were motivated to write this chapter due to the increased interest in this
sector since FL emerged in 2016, with a wide variety of applications, challenges,
and concerns related to this new paradigm. As a result, there is a paucity of recent
survey papers and preprints that discuss the FL field from various viewpoints.
AbdulRahman et al.’s (2020) research on federated learning for mobile-edge net-
works highlights challenges in privacy, security, resource management, and commu-
nication overhead. Additionally, it shows some uses of FL for edge networks. Various
FL settings, including horizontal FL, vertical FL, and compound transfer learning,
were categorized and described by AbdulRahman et al. (2020) and Aledhari et al.
(2020) according to the characteristics of the data distribution. Focusing on these
four key areas—communication, system heterogeneity, statistical heterogeneity,
and privacy—can lead to more effective federated learning (FL) implementations.
Wei et al. (2020) concentrate on implementation issues and its current techniques.
Wireless communication was highlighted by Konečný et al. (2016) as a potential
application area for FL.
Given the rapid growth, wide range of applications, and significant impact of FL
in various fields of study, it is surprising that there has not been a literature review
that covers all the bases regarding the modeling, implementation, and technical
Enabling Federated Learning in the Classroom 39

aspects of FL in terms of sociotechnical aspects, details, and deployment and directs

researchers to relevant contributions. This fact prompted us to conduct comprehen-
sive survey research, identify key issues, and develop a survey that incorporates the
new taxonomy. Therefore, to overcome the gaps that occur regarding the form of
implementing integrated learning in the classroom, the integration of artificial intel-
ligence in education holds great promise for personalized learning and enhanced
educational experiences by leveraging student data centers and concerns regarding
data privacy and security by facilitating learning spaces with a federated learning
approach, offering a secure and collaborative approach to leveraging AI in the class-
room with sociotechnical ecosystems in educational practice.

SMART SOCIOTECHNICAL ECOSYSTEM

THROUGH FEDERATED LEARNING MODEL
In general, the integration of artificial intelligence in educational practices has great
potential for personalized learning and enhancing more meaningful educational
experiences Al (Yakin et al., 2023). It cannot be denied that surfing using tech-
nology raises concerns about privacy and data security, which results in significant
challenges (Bhutoria, 2022). Federated learning is emerging as a promising solution,
enabling the development of collaborative AI in the classroom while maintaining
the privacy of student data while interacting with this AI. There are several benefits
of FL in learning, such as that with FL, you can overcome data privacy problems by
storing student data on their devices, such as smartphones, tablets, or laptops. This
decentralization eliminates the need to share raw data with a central server, thereby
reducing the risk of breaches, misuse, and dissemination of personal data (Yuan
et al., 2023).
According to Southworth et al. (2023), collaborative FL by integrating AI can
foster collaboration between institutions or classrooms by allowing them to partici-
pate in model training without compromising the privacy of existing student data.
Each device automatically trains a local model based on its data, and only model
updates, not raw data, are shared or disseminated. This collaboration facilitates the
development of more robust and generalizable AI models that can learn from diverse
educational contexts. The most popular FL capabilities are training models on local
data and facilitating personalized learning experiences in the learning space, offline
or online. Therefore, by utilizing individual student data, customized models can be
created to meet specific needs and learning styles (Ji et al., 2021). Based on these
conditions, adjusting educational content, suggesting relevant learning resources, and
providing differentiated feedback can be maximized. Traditional AI models often
require large computing resources, creating a bottleneck for smaller institutions or
those with limited resources. FL distributes the training workload across participat-
ing students’ devices, reducing resource burden, making AI-supported education
more accessible, and creating learning convenience (Dhananjaya et al., 2024).
To better understand the role of FL (Han and Zhang, 2020), it is necessary to
study its interaction with the broader sociotechnical ecosystem of AI integration in
education. This ecosystem includes a variety of stakeholders, including students,
teachers, developers, and policymakers, who interact with and influence technology
40 Artificial Intelligence Using Federated Learning

in the context of existing educational practices and social norms. Studies highlight
the importance of considering the pedagogical implications of AI integration (Hess
et al., 2021). Integrating FL successfully requires careful alignment with existing
curricula and teaching philosophies, ensuring that AI tools complement and enhance,
not replace, traditional pedagogy. In addition, the human element remains important,
where teacher training and support are essential to efficiently maximize the potential
of FL in the classroom (Lazarus et al., 2023).
Accelerated AI and federated learning provide promising avenues in education
while addressing data privacy concerns. However, it is important to acknowledge
the existing challenges and consider FL within the broader sociotechnical ecosys-
tem. By fostering collaboration, addressing ethical considerations, and thoughtfully
integrating FL into existing pedagogical practices, the educational community can
harness FL’s unique potential to create personalized and engaging learning experi-
ences for all students in higher education (Bienkowski et al., 2012).
The digitalization and development of Mixverse in the educational landscape are
undergoing transformative change as educators and institutions leverage innovative
technologies to revolutionize learning and teaching practices. The rapid advance-
ment of artificial intelligence has opened the door to new applications in various
sectors, including education. Although AI’s potential for personalized learning
and improving educational experiences has been widely recognized, effectively
integrating AI into the classroom environment still presents ongoing challenges.
Artificial intelligence has great potential to personalize learning experiences, meet
individual student needs, and provide educators with valuable insights into student
progress. However, integrating AI into a complex and dynamic classroom environ-
ment requires careful consideration of factors beyond the purely technical aspects.
Existing research primarily focuses on the technical aspects of applying AI in edu-
cational settings, with limited exploration of the broader sociotechnical ecosystem.
These discovered gaps require a deeper understanding of the social and technical
factors that influence the successful integration of AI in the classroom, considering
the diverse stakeholders and their interactions (Lazarus Cowin et al., 2023).
This chapter aims to address this critical gap by exploring the sociotechnical
ecosystem surrounding the integration of AI in educational practices, particularly in
higher education. We will examine the social and technical factors that influence the
successful implementation of AI-powered learning in the classroom environment,
with a focus on the interactions between technology, pedagogy, and human actors.
More and more research is being conducted to investigate the potential applications
of AI in education. Research has explored AI-powered tutoring systems, personalized
learning platforms, and intelligent feedback mechanisms, showing promising results
in various educational contexts. Experts emphasize the importance of considering
the pedagogical implications of AI integration. Inderawati et al. (2024) argue that
successful implementation requires careful alignment with the existing curriculum
and teaching philosophy. Additionally, the human element remains important, and
Aeni et al. (2024) highlight the need for educators to have the necessary skills and
knowledge to effectively utilize and integrate AI tools into their teaching practices.
Additionally, the research underscores the social complexity involved in AI inte-
gration (Sadek et al., 2024) and raises concerns about potential biases embedded in
Enabling Federated Learning in the Classroom 41

AI algorithms, necessitating responsible development and implementation practices.

Additionally, Le-Nguyen and Tran (2024) emphasize the importance of addressing
ethical considerations around data privacy, student agency, and the potential for AI
to exacerbate existing educational disparities. By exploring the sociotechnical eco-
system, this research aims to bridge the gap between the technical development of
AI and its successful integration in the complex social settings of the classroom. By
examining the interactions between technology, pedagogy, and human actors, we
seek to contribute valuable insights to guide the responsible and effective implemen-
tation of AI-powered learning for the benefit of all stakeholders in the educational
process.
This chapter provides insights into the integration of blended learning in educa-
tional practice from a sociological perspective. Although blended learning prom-
ises to revolutionize education, its successful implementation requires a different
understanding of social dynamics in the classroom. By combining theory and socio-
logical frameworks, educators and policymakers can address challenges and harness
the benefits of AI to drive academic excellence in diverse learning environments.
Although they cover important and fundamental issues in many FL subjects and
research domains, such as resource management, privacy and security, and applica-
tion areas, as well as core system models and design, George and Wooden’s paper
(2023) was published at the same time as this survey, but the approach is differ-
ent. We anticipate that this chapter will provide a comprehensive picture of existing
issues, categorize them, and help researchers develop effective strategies for promot-
ing a variety of new technologies and current topics. The following are the main
points of this work compared to previous surveys.

1. We detail the development of students’ knowledge construction with arti-

ficial intelligence-based FL approaches in the classroom, examine FL sub-
jects and research fields in detail, classify the contributions and efforts of
FL paradigms currently trending in research and industry, and provide a
detailed overview and core analysis of the model FL and system design,
covering important technical aspects. Additionally, we discuss the difficul-
ties and interesting unanswered questions that lay the foundation for future
FL solutions in the educational landscape.
2. To cover all domains that offer FL techniques so far, we create a taxonomy of
FL application areas in the field of education and analyze sociological aspects,
which are of course very different from researchers in the fields of Internet of
Things, edge computing, networking, robotics, the world of networks, models,
recommendation systems, cybersecurity, online classes (learning manage-
ment system/LMS), wireless communications, and electric vehicles.
3. We go further by reviewing the literature to determine whether important
contributions to the FL paradigm have highlighted the complex interactions
between technology and social structure.
4. We classify FL as a term learning approach with optimization methods
based on the objective function and calculated parameters, and we provide
an in-depth evaluation of the recommended resource management strate-
gies for FL scenarios.
42 Artificial Intelligence Using Federated Learning

DECENTRALIZATION OF EDUCATION THROUGH

AI FOR THE SOCIOTECHNICAL ECOSYSTEM
An innovative method for protecting student information while incorporating arti-
ficial intelligence into classroom activities is presented by federated learning (Tan
et al., 2022). By tackling current issues and encouraging teamwork, FL may revolu-
tionize classroom instruction, opening the door to individualized learning and pro-
viding teachers with innovative resources to meet the needs of their students from
all backgrounds. Future research and development efforts could position FL to play
a pivotal role in the ethical and successful integration of AI into classrooms, paving
the way for more engaging and tailored teaching for all students.
Integrating new technologies like artificial intelligence is currently driving a fun-
damental transition in the field of education. But getting students to use technology
in class is only half the battle. To guarantee successful technological integration,
a sociotechnical approach is necessary, which highlights the intricate interplay
between social and technical elements and algorithms. The sociotechnical theory
posits that the success of any system including educational systems depends on
the degree to which its social and technical components are interdependent on one
another. Acknowledging the interplay between social elements like pedagogy, stu-
dent demands, and school culture and technological elements like the purpose, appli-
cations, and constraints of educational technology (e.g., platforms driven by artificial
intelligence) is essential in educational taxonomy. To integrate technology smoothly
and successfully, understanding these interconnections is crucial (Linderoth et al.,
2024).
Educators must be provided with the information and training to make success-
ful use of technology, according to multiple studies. This calls for an appreciation
of technology’s possibilities, its incorporation into the educational process, and the
resolution of any obstacles that may arise for the benefit of students (Denny et al.,
2024). Gill et al. (2024) conducted a study to explore optimal methods for integrating
technological tools into traditional educational structures. Adaptations to teaching
methods may be necessary, for instance, if certain technologies necessitate a change
to a student-centered learning style.
Other studies look at serious problems like algorithmic bias, student agency, and
data privacy. For technology to be used fairly and ethically in education, it must be
developed and used responsibly. In addition, research by Kaliraj et al. (2024) investi-
gated ways technology might help close achievement disparities and make education
accessible to everyone. This necessitates attending to details like the availability of
devices and dependable internet connections for students. Technology offers numer-
ous advantages in education when a sociotechnical approach is used. To ensure that
technology is seamlessly integrated into the classroom, taking into account both social
and technical factors is crucial. This will lead to an engaging and productive learning
environment for students, as technology can be utilized to create personalized learn-
ing paths that cater to each student’s unique needs and learning style. By doing so, we
can effectively support a diverse range of students. According to research (Goel et al.,
2024), teachers can spend more time fostering meaningful student engagement and
cooperation if they use technology to streamline administrative work.
Enabling Federated Learning in the Classroom 43

There are no longer clear boundaries between the two phases of learning: the
classroom and the workplace (Southworth et al., 2023). The average person today
has more knowledge than they can process, and the workforce of the future will
need to know more than that. Promoting high-quality educational opportunities
across life’s transitional contexts—from the family home, classroom, workplace, and
beyond—is critical to fostering lifelong learning. Rather than starting with a certain
level of education, professionals need to find ways to incorporate learning into their
daily work. Learning also occurs in many environments, such as families, clubs, and
online communities, and occurs at all ages and in almost every occupation. The find-
ings from these specific cases should inform more generalizable learning theories
and more creative and perceptive systems, practices, and evaluations in a variety of
fields. With a focus on continuous learning, the benefits of learning in the classroom,
in the community, at home, and work can all be combined.
Increasing professional specialization, market changes (local and global are no
longer distinguishable), and new technologies for information sharing, communi-
cation, and collaboration, made possible by Internet technologies and specialized
applications, are characteristics of today’s society (Abulibdeh et al., 2024). The
skills taught in school, such as reading, writing, and arithmetic, are not sufficient for
situations like this. Some examples of new capabilities include the ability to work
remotely, communicate effectively in diverse teams, build consensus, adapt knowl-
edge artifacts, address problems with unclear or missing specifications, and utilize
online and specialized collaboration.
According to Ali et al. (2024), the following are some characteristics of the socio-
technical ecosystems that have emerged throughout the past 10 years. The deep
and lasting digitalization and transformation of our time are rooted in social and
cultural change as well as technology. Rather than technology being the primary
driver of change in complex situations, Sriram (2024) argues that human behavior
and social organization undergo gradual change. He said that this includes not only
hardware, networks, and software but also people, processes, regulations, laws, and
the movement of raw materials and finished products, as well as many other com-
ponents. In addition, these conditions require the co-design of social and technical
systems using models and concepts that consider the artifacts and the social context
in which the system will be used. Meta-design between AI, social interaction, and
federated learning is an important part of this system because it empowers users
to shape and adapt the technical system to their own needs. As we know, this system
is designed for humans, not just one person. They must be able to support both indi-
viduals and groups, but groups achieve their goals through the combined efforts of
all members and involve collaboration. Therefore, the question of how to create an
environment that encourages individual and social efforts without disrupting either
is of paramount importance.
According to Bednar and Welch (2020), we need new perspectives and methods
to address the fundamental problems associated with distributed intelligence and
the development of sociotechnical systems to help people live better and more ful-
filling lives, so we researched to find ways of exploiting the potential of new tech-
nology based on a widespread and reliable computing environment and increasing
people’s proficiency in using technology in the mixverse era. Rather than conveying
44 Artificial Intelligence Using Federated Learning

pre-digested information, new media and technologies can play an important role
in facilitating meaningful activities, social debate and discussion, stakeholder
understanding, and the formulation and resolution of real problems. Based on this
worldview, the following sociotechnical arrangements are necessary to facilitate
self-directed learning and learning communities in higher education. These systems
should facilitate the enhancement of individual and collective knowledge; this means
that they not only help individuals when they work alone but also when they collabo-
rate with others and are members of several groups simultaneously, whether through
digital interactions or interacting with objects or people in the real world. This seems
to be an excellent strategy to improve the learning experience.
Improvisation, evolution, and creativity are nothing extra in a world where noth-
ing is guaranteed. Lecturers must use it as an opportunity to find better and original
answers to a problem. Sociotechnical arrangements that enable new collaborative
designs are needed to empower self-directed learners and reinvigorate social innova-
tion in learning communities. To empower users to take on the role of designers and
unleash their creativity, meta-design outlines goals, methods, and procedures (Tabo,
2020). AI-enabled technologies can support self-directed learning by shifting the
focus from mere consumption of information to active participation in its production.
AI-based federated learning goes beyond user-centered and participatory learning
design in many ways and is encouraged to share ideas and goals.

METHODOLOGY
This research used quantitative methods. We used Google Forms to create our online
surveys. This survey consisted of five parts. In the first part, we tried to collect some
basic information from participants, including basic demographics such as student
age, gender, and self-reported level of technology competency.
A Likert scale consisting of 5 points is used to assess technology skills such
as: strongly agree, agree, somewhat agree, disagree, and disagree. In the second
part of the chapter, we look at how people understand the basics of AI (such as
learning using AI bots); how they utilize AI in learning (such as accessing infor-
mation); and how they process, share, and discuss information. Asking whether
participants were aware that AI and deep learning are widely discussed in the edu-
cational community, the third subsection sought to explore the understanding of
AI as a topic in learning. In the fourth part, we asked students to rate their feelings
and attitudes towards artificial intelligence, considering their answers to the follow-
ing questions: (1) how they view AI about humans and society; (2) how they feel
towards artificial intelligence as a tool for socializing information and knowledge;
(3) how they feel about the development of AI in the federated learning model; (4)
how they feel about AI regarding its potential to improve learning outcomes and
its integration into the teacher professional education curriculum. Using five differ-
ent 5-point Likert scales, these sections were scored: (5) strongly agree, (4) agree,
(3) somewhat agree, (2) disagree, and (1) strongly disagree. The five components
included questions about respondents’ opinions about the use of artificial intelli-
gence in social interactions in the classroom and their hopes for future education in
this area Figure 3.1.
Enabling Federated Learning in the Classroom 45

FIGURE 3.1 Playing AI on federated learning model.

STATISTICAL ANALYSIS
The results of data analysis, namely the average value and standard deviation of continu-
ous variables, are displayed. Data presented as percentages and frequencies are known
as categorical variables. We described “strongly agree” and “agree” as agreement for the
descriptive statistics, and “agree” and “strongly agree” as disagreement. The results of
this research data were processed using the Statistical Package for Social Sciences (SPSS
version 20) used for data entry and analysis. Data was analyzed using descriptive statistics
to obtain percentages, frequencies, standard deviations, and means. The reliability and
validity tests were determined by calculating Cronbach’s alpha, which produced 0.82.

RESULTS: SURVEY RESPONDENTS

A total of 43 respondents from professional teacher education took the time to
respond and were considered for analysis. Upon completion of the study, precise
data will be accessible concerning the enrollment figures and demographic profile
of professional teacher education (PPG) students for the 2023 and 2024 academic
years. Preliminary findings indicate that there were 20 first-year teachers and 13 sec-
ond-year students, with a gender distribution of approximately 35% male and 65%
female. The majority of participants were aged 22 to 25, with 24 respondents falling
within this age range and 19 specifically being 22 to 25 years old. In terms of device
ownership, all respondents owned smartphones and laptops. Most respondents had
prepaid internet subscriptions using smartphone data to access the internet.

OVERALL PERCEPTION AND CATEGORY OF RESPONDENTS’

RESPONSES TOWARD SOCIOTECHNICAL AI
Studying new technology courses in the first year and studying design thinking in the
second year, this research intends to investigate how professional teacher education
students view AI in education with a federated learning model. AI-based education
46 Artificial Intelligence Using Federated Learning

is strongly supported by the results of this poll. Learning with AI is often seen as
preferable by students in higher education, and they state that taking this class is
their first experience interacting with AI and getting to know AI.
This leads the researchers to the confident conclusion that AI learning with the FL
model is very attractive for professional teacher education students at Al Asyariah
Mandar University. Even though inadequate internet conditions due to having to
prepare quotas do not provide a big obstacle, this has a positive impact on the atomic
habits of those who adapt to AI. This activity certainly uses creative ways of teach-
ing and learning so that the educational process continues and is meaningful. In
many cases, students perceive learning with AI integration as more comfortable and
safer. In conclusion, the findings of recent research support students’ positive atti-
tudes toward AI learning with the federated learning model described in Table 3.1.
The results of survey data analysis regarding respondents’ opinions regarding
the integration of AI with federated learning in the sociotechnical ecosystem model
described in Table 3.1 show that students generally have a positive attitude towards
artificial intelligence in education, especially with federated learning. In both state-
ments, students demonstrated a reasonable level of understanding regarding the ben-
efits of AI in education (Statement 1) and their belief that AI-driven learning tools
are becoming more common (Statement 2). As stated in Statement 3, they enjoy
leveraging AI with FL, and as stated in Statement 4, they recognize its potential
in accessing learning materials. Many people feel comfortable disclosing personal
information to AI systems (Statement 5). Students demonstrated a greater level of
comfort in terms of AI information processing, while they generally felt comfortable
using AI tools (Statement 6). Statement 8 stated that students had a positive impres-
sion of sharing and debating material in class using AI tools. Personalized learning
(Statement 9) and collaboration (Statement 10) are two areas that this research sug-
gests could be beneficial.
In the “mixverse era” (Statement 11), students recognize the importance of AI
and deep learning, and they feel comfortable discussing AI in class (Statement 12).
According to Statement 13, socio-technical skills can be improved by incorporating
AI-based learning into the curriculum. The broader impact of AI on our ability to
understand the universe may require further investigation (Statement 14). Finally,
the survey shows that students support the development of AI in federated learning
(Statement 18), believe that AI can improve learning outcomes (Statement 19), feel
comfortable relying on AI for learning (Statement 17), have positive perceptions of
the role of AI in society (Statement 16), and strongly support integrating AI into
teacher education (Statement 20). The results of this research show a bright future
for educational technology that utilizes artificial intelligence in the sociotechnical
ecosystem, as shown in Figure 3.2.
Using the descriptive statistics in Table 3.2, this study investigated student senti-
ment regarding educational approaches involving federated learning and artificial
intelligence. By providing a brief overview of the data, descriptive statistics help us
understand the data’s central tendency, variability, and the shape of the distribution
of students’ answers. There is only one variable in Table 3.2, and the descriptive
statistics aimed at this variable might be a survey question asking students’ opinions
about FL and AI models. This question uses a scale with a range of 39 points, which
TABLE 3.1
Student Survey

Enabling Federated Learning in the Classroom

Std.
Statements N Minimum Maximum Mean Deviation
Part 1: Understanding AI basics through FL model (5 items) 43 3.00 5.00 4.4651 .59156
1. I understand the functions and benefits of artificial intelligence (AI) in education through the federated learning model.
2. I interact with AI-powered tools in class and outside of class to access information about education or teaching. 43 3.00 5.00 4.2093 .67465
3. I enjoy learning to use AI with federated learning in class and outside of class. 43 2.00 5.00 4.3721 .65550
4. I believe that AI can be used to access information for learning purposes (e.g., searching websites, and finding 43 3.00 5.00 4.2791 .66639
relevant resources).
5. I feel comfortable sharing information and interacting socially with AI-powered learning tools. 43 2.00 5.00 4.2558 .84777
Part 2: Utilizing AI in learning (5 items)
6. I feel comfortable using AI-powered tools in the learning process. 43 2.00 5.00 3.9535 .78539
7. I believe AI helps process the information to learn. 43 3.00 5.00 4.3023 .67383
8. I believe that it is easy to share information with others through AI-powered tools (e.g., online discussion forums) 43 3.00 5.00 4.3023 .63751
and discuss in the classroom.
9. I find it useful to use AI to personalize the learning experience for different students. 43 3.00 5.00 4.3953 .62257
10. I believe AI can be a valuable tool for collaborating and learning with others. 43 3.00 5.00 4.4186 .62612
Part 3: AI as a learning topic (5 items)
11. I believe AI and deep learning essential in the mixverse era. 43 3.00 5.00 4.1860 .69884
12. I am comfortable carrying out discussion activities using AI in education. 43 2.00 5.00 4.4419 .70042
13. I believe that AI-based learning as part of the regular curriculum is useful for improving socio-technical skills. 43 2.00 5.00 4.3256 .74709
14. Learning with AI can help students better understand the world around them. 43 2.00 5.00 3.5581 .82527
15. Students should be involved in discussions about the use of AI in education. 43 3.00 5.00 4.3721 .61811
Part 4: Attitudes toward AI (5 items)
16. I see the role of AI as very effective in relation to humans and society in general. 43 3.00 5.00 4.2791 .62965
17. I feel comfortable relying on AI tools to learn and disseminate information and knowledge. 43 3.00 5.00 4.3488 .68604
18. I agree with the development of AI in a federated learning model (where the data remains on the user’s device). 43 3.00 5.00 4.1163 .69725
19. I feel the potential of AI to efficiently improve student learning outcomes. 43 3.00 5.00 4.1628 .75373

47
20. I believe that the integration of AI into teacher professional education curricula is very important. 43 3.00 5.00 4.3488 .68604
Valid N (listwise) 43
48 Artificial Intelligence Using Federated Learning
FIGURE 3.2 Histogram and normal P-P plot.
Enabling Federated Learning in the Classroom 49

TABLE 3.2
Description of Statistics
Std.
N Minimum Maximum Mean Deviation Skewness Kurtosis
Std. Std.
Statistic Statistic Statistic Statistic Statistic Statistic Error Statistic Error
AI-FL model 43 60 99 85.09 9.633 –.554 .361 –.194 .709
Valid N (listwise) 43

TABLE 3.3
Test Statistics
Algorithm Test AI-FL model
Chi-square 8.884a
DF 22
Asymp. sig. .994

Twenty-three cells (100.0%) have expected frequencies less than 5.

The minimum expected cell frequency is 1.9.

is indicated by a minimum score of 60 and a maximum score of 99. With this scale
being a Likert scale, survey statements regarding FL and AI models can cover an
alternative range from strongly disagree (60) to strongly agree (99). Table 3.2 shows
a mean score of 85.09 and a standard deviation of 9.633. Students generally seem to
have a positive view of FL and AI models. However, there is a lot of variation in these
sentiments, as seen in the standard deviation. In other words, some children have a
more optimistic viewpoint than others. Kurtosis (.361) and skewness (−.554) values:
A slight negative skew indicates a slight bias towards higher scores, which could be
an indication of a positive skew attitude. If the kurtosis value is close to 3, it means
the distribution is almost normal.
Data results on the chi-square test (Table 3.3) summarize the chi-square test sta-
tistics applied to the AI variable with the federated learning model on the social-
technical ecosystem (likely representing student responses to a survey about attitudes
towards AI and the federated learning model). It is known that the data shows that
the chi-square test statistic (8.884a) is relatively low. The degrees of freedom (DF)
are 22, indicating the number of categories used to classify student responses.
Asymptotic significance (asymp. sig.) is a very high value (0.994), exceeding the
common threshold of 0.05 for statistical significance.
The findings of the reliability analysis for the AI variable with the federated
learning model on the social-technical ecosystem, which uses Cronbach’s alpha, are
displayed in Table 3.4. This variable is a collection of 20 survey questions intended to
measure how students feel about educational approaches that use federated learning
50 Artificial Intelligence Using Federated Learning

TABLE 3.4
Reliability Statistics
Cronbach’s Alpha N of Items
.943 20

and artificial intelligence. Survey data results with 20-item questions measuring
opinions on AI and FL models and Cronbach’s alpha show how well these questions
measure the same basic concepts. The very high result of 0.943 exceeds the gener-
ally recognized criterion of 0.7 for satisfactory internal consistency. This suggests a
strong correlation between survey items and their ability to measure cohesive ideas.
With such a high alpha value, researchers can be confident that this research variable
is a reliable indicator of student views.
The research results showed that there was greater confidence that the survey
could reflect students’ true opinions.We can now more accurately identify instances
where students’ beliefs differ across categories or contexts. Although a high alpha
implies that the survey is internally consistent, it does not guarantee content valid-
ity. The results of this research survey have a very high level of internal consistency,
as indicated by a Cronbach’s alpha value of 0.943. Due to its high alpha value, this
instrument is overall reliable for measuring students’ opinions about the use of FL
and AI in the classroom.

DISCUSSION
Table 3.1 examines a student survey asking about their knowledge, use, and feelings
towards AI in education, specifically as it relates to federated learning. According to
the survey results presented in Table 3.1, students who have received FL instruction
demonstrated a moderate level of comprehension regarding the characteristics and
advantages of artificial intelligence in the educational setting (Statement 1, mean =
4.47). They also believe that the use of AI-powered tools for educational purposes
is increasing, both inside and outside the classroom (Statement 2, mean = 4.21). The
next statement claimed that using FL and AI was a pleasant experience for students,
according to Statement 3 (mean = 4.37). Furthermore, from the data, it is known
that there is a high consensus regarding AI’s ability to access learning material
(Statement 4, mean = 4.28). Additionally, factors influencing the ease with which
individuals share data and interact with AI-powered learning tools vary (Statement 5,
mean = 4.26).
Statement 7, mean = 4.30, indicates a higher level of comfort for students regard-
ing the involvement of AI in processing learning information, while Statement 6,
mean = 3.95, indicates that students feel comfortable using AI tools overall. Statement 8,
mean = 4.30, indicates that students like sharing information through AI-powered
tools and discussing it in class. According to the data results in Statement 9, the
average score is 4.39, which shows that the possibility of AI for individualized edu-
cational experiences is recognized. According to students in the teacher education
Enabling Federated Learning in the Classroom 51

profession, AI has the potential to improve learning and collaboration (Statement 10,
mean = 4.42). In Statement 11, the mean is 4.19, recognizing the importance of AI
and deep learning in the “mixverse era” (possibly referring to hybrid physical-virtual
reality settings). Statement 12, mean = 4.44, indicates that students feel comfort-
able talking about AI in class. Then the poll shows that socio-technical skills can
be improved by incorporating AI-based learning into the curriculum, which can be
seen in Statement 13 (mean = 4.33).
A more thorough investigation of this subject may be needed, although there is an
encouraging trend in the use of AI to learn about the world (Statement 14, mean =
3.56). Then, Statement 15, mean = 4.37, shows that students support participation in
class debates regarding the use of AI in the classroom. Most of the students expressed
positive views regarding the usefulness of AI for humans and society (Statement 16,
mean = 4.28). Based on Statement 17, the average score of students’ self-confidence
in using AI technology for learning and disseminating knowledge is 4.35.
There is a consensus on the need to develop AI in FL, where user data is stored
on the device, according to Statement 18 (mean = 4.12). Statement 19, mean = 4.16,
shows that the public has a positive impression of AI’s ability to improve learning
outcomes. Then, according to Statement 20 (mean = 4.35), this research shows that
there is great support for incorporating AI into teacher professional education. Based
on the survey, it is known that students generally have a favorable opinion of AI in
education. This is especially true when it comes to federated learning. The positive
impact of AI on education, teamwork, and individualization has been recognized
by students. This poll opens the door to further research and the potential use of
AI-powered learning aids in the classroom, although this may require further inves-
tigation in certain areas, such as the global impact of AI.
Additional details regarding the appropriate choice of survey questions and
answers are needed for a more thorough interpretation, as this analysis relies on a
single variable. Furthermore, descriptive statistics by themselves cannot reveal group
differences and causal relationships. Students’ perspectives on the use of FL and AI
models in the classroom can be better understood with the use of descriptive statis-
tics (Tables 3.1 and 3.2). Although every respondent’s answers are different, the data
shows an optimistic future. We can look at the relationships between these attitudes
and other student traits or learning outcomes in future studies. We can learn more
about students’ views on the pros and cons of using FL and AI models in the class-
room by looking at the survey questions themselves and using inferential statistics.
This chapter provides insights into the integration of blended learning in educa-
tional practice from a sociological perspective. Although blended learning prom-
ises to revolutionize education, its successful implementation requires a different
understanding of social dynamics in the classroom. By combining theory and socio-
logical frameworks, educators and policymakers can address challenges and harness
the benefits of AI to drive academic excellence in diverse learning environments.
Although they cover important and fundamental issues in many FL subjects and
research domains, such as resource management, privacy and security, and applica-
tion areas, as well as core system models and design, Yu et al. (2023) and Sartori and
Theodorou (2022) were published at the same time as this survey, but the approach
is different. We anticipate that the suggested survey will provide a comprehensive
52 Artificial Intelligence Using Federated Learning

picture of existing issues, categorize them, and help researchers develop effective
strategies for promoting a variety of new technologies and current topics.
The following are the main points of this work compared to previous surveys.
(1) We detail the development of students’ knowledge construction with artificial
intelligence–based FL approaches in the classroom, examine FL subjects and
research fields in detail, classify the contributions and efforts of FL paradigms cur-
rently trending in research and industry, and provide a detailed overview and core
analysis of the model FL and system design, covering important technical aspects.
Additionally, we discuss the difficulties and interesting unanswered questions that
lay the foundation for future FL solutions in the educational landscape. (2) To cover
all domains that offer FL techniques so far, we create a taxonomy of FL application
areas in the field of education and analyze sociological aspects which are of course
very different from researchers in the fields of Internet of Things, edge comput-
ing, networking, robotics, the world of networks, models, recommendation systems,
cyber security, online classes, wireless communications and electric vehicles. (3) We
go further by reviewing the literature to determine whether important contributions
to the FL paradigm have highlighted the complex interactions between technology
and social structure. (4) We classify FL as a term learning approach with optimiza-
tion methods based on the objective function and calculated parameters, and we
provide an in-depth evaluation of the recommended resource management strategies
for FL scenarios.
Based on the results of this research, we realize that the inseparability of social
and technological components in work is important from a socio-technical systems
perspective, as has been done by Farrow (2023) and Schoenherr (2024), which is
similar to this research, in which humans, tasks, technology, AI, physical space, and
organization are the five pillars underlying the work system model. Workflow refers
to the interactions between various components of a working system that develop
over time as individuals, aided by tools and technology, perform activities in a physi-
cal setting within an educational framework. The sociotechnical system with the
federated learning model is only one part where the application of AI technology
influences changes in the relationship between elements of the work system in the
Merdeka class. The new system’s integration will significantly alter how teacher pro-
fessional education students interact with technology, tasks, and their peers within
the learning environment. This involves ensuring that technology seamlessly fits
into their workflow, considering factors like task sequencing and the timing of tech-
nology usage. Although the results of the data from respondents did not claim that
negative consequences (such as distraction and increased workload) can result from
failure to integrate technology into the workflow during the learning process, good
outcomes (such as acceptance and utilization) can occur when technology is imple-
mented. Therefore, AI must be built to complement, not hinder, workflows in edu-
cation by integrating it with other system components. For AI to be successfully
integrated into these workflows, all aspects of the work system and the various levels
and dimensions of workflow integration need to be considered.
This research is different from other research in that implementing an AI system
not only emphasizes the development of AI technology but also confirms it with stu-
dents’ real-world environments, such as information obtained from AI that remains
Enabling Federated Learning in the Classroom 53

confirmed with other scientific sources. Of course, the results of this study refute the
arguments of Salwei and Carayon (2022), which state that artificial intelligence is
often created in a vacuum, ignoring the real world in which it will be used. Artificial
intelligence has the potential to improve and support the learning process; however,
the success of this technology depends on the acceptance of students and lecturers
who teach with AI (Muthmainnah et al., 2023). Especially in decision-making and
criticizing information, this happens. A lack of confidence and reluctance to utilize
technology can arise from unclear communication about both the technology and its
recommendations.

CONCLUSION
Our definition and operationalization of the sociotechnical scope that focuses on
AI technology with federated learning models shows significant promise. Based on
these results, we can better understand masking in general and find more socially
and technically oriented methods in the learning process in Merdeka classes or the
independent curriculum (KURMA). We have briefly reviewed the capabilities and
limitations of various sociotechnical approaches to addressing this problem as a
starting point. This work is purely the result of the opinions of professional teacher
education students. We base our considerations on the premise that AI systems are
getting a positive response and developing as part of the digitalization of education.
Therefore, the application of AI in the classroom and the social and sociotechnical
interactions formed with the federated learning model requires human intervention.
Additionally, even if AI systems develop further in the future, we still think that
humans will play an important role. Because our complementary approach views
humans and technology as fundamentally different, we operate on this premise.
Therefore, they did not compete to see which of the two had quantitatively compara-
ble superior abilities. The reality is that their different strengths work better together.
Computers excel in computing, but humans excel in thinking, critical thinking, sys-
tems thinking, and collaboration. Therefore, no matter how good human intelligence
or technological capabilities are, an intelligent human–technology combination will
always outperform both, as has been proven through this research. When it comes
to AI systems, we strongly believe that this is especially the case in fields where
understanding and responsibility are critical human traits, where combined human
AI systems will work best. Rather, it is a human need for the conditions necessary
for education to have a high level of involvement and higher-order thinking skills.
The reason is that by involving technology, showing initiative, showing dedication,
or accepting responsibility cannot be done simply by direction or assignment, but
proper work design is essential so that human contribution can be motivated authen-
tically by the design of socio-technical systems, i.e., the level of allocation of human
functions to machines in the mixverse class.

REFERENCES
AbdulRahman, S., Tout, H., Ould-Slimane, H., Mourad, A., Talhi, C., & Guizani, M. (2020).
A survey on federated learning: The journey from centralized to distributed on-site
learning and beyond. IEEE Internet of Things Journal, 8(7), 5476–5497.
54 Artificial Intelligence Using Federated Learning

Abulibdeh, A., Zaidan, E., & Abulibdeh, R. (2024). Navigating the confluence of artificial
intelligence and education for sustainable development in the era of industry 4.0:
Challenges, opportunities, and ethical dimensions. Journal of Cleaner Production,
140527.
Aeni, N., Khang, A., Al Yakin, A., Yunus, M., & Cardoso, L. (2024). Revolutionized teach-
ing by incorporating artificial intelligence chatbot for higher education ecosystem. In
AI-Centric Modeling and Analytics (pp. 43–76). CRC Press.
Aledhari, M., Razzak, R., Parizi, R. M., & Saeed, F. (2020). Federated learning: A survey on
enabling technologies, protocols, and applications. IEEE Access, 8, 140699–140725.
Ali, M., Wood-Harper, T., & Wood, B. (2024). Understanding the technical and social para-
doxes of learning management systems usage in higher education: A sociotechnical
perspective. Systems Research and Behavioral Science, 41(1), 134–152.
Al Yakin, A., Khang, A., & Mukit, A. (2023). Personalized social-collaborative iot-symbiotic
platforms in smart education ecosystem. In Smart Cities (pp. 204–230). CRC Press.
Bednar, P. M., & Welch, C. (2020). Socio-technical perspectives on smart working: Creating
meaningful and sustainable systems. Information Systems Frontiers, 22(2), 281–298.
Bhutoria, A. (2022). Personalized education and artificial intelligence in the United States,
China, and India: A systematic review using a human-in-the-loop model. Computers
and Education: Artificial Intelligence, 3, 100068.
Bienkowski, M., Feng, M., & Means, B. (2012). Enhancing teaching and learning through
educational data mining and learning analytics: An issue brief. Office of Educational
Technology. US Department of Education.
Čolaković, A., & Hadžialić, M. (2018). Internet of Things (IoT): A review of enabling tech-
nologies, challenges, and open research issues. Computer Networks, 144, 17–39.
Denny, P., Gulwani, S., Heffernan, N. T., Käser, T., Moore, S., Rafferty, A. N., & Singla,
A. (2024). Generative AI for education (GAIED): Advances, opportunities, and chal-
lenges. arXiv preprint arXiv:2402.01580.
Dhananjaya, G. M., Goudar, R. H., Kulkarni, A., Rathod, V. N., & Hukkeri, G. S. (2024). A
digital recommendation system for personalized learning to enhance online education:
A review. IEEE Access, 12, 34019–34041. doi: 10.1109/ACCESS.2024.3369901
Donnelly, R. (2017). Blended problem-based learning in higher education: The intersec-
tion of social learning and technology. Psychosociological Issues in Human Resource
Management, 5(2), 25–50.
Farrow, R. (2023). The possibilities and limits of XAI in education: A socio-technical per-
spective. Learning, Media and Technology, 48(2), 266–279.
George, B., & Wooden, O. (2023). Managing the strategic transformation of higher education
through artificial intelligence. Administrative Sciences, 13(9), 196.
Gill, S. S., Xu, M., Patros, P., Wu, H., Kaur, R., Kaur, K.,. . .& Buyya, R. (2024). Transformative
effects of ChatGPT on modern education: Emerging era of AI chatbots. Internet of
Things and Cyber-Physical Systems, 4, 19–23.
Goel, P. K., Singhal, A., Bhadoria, S. S., Saraswat, B. K., & Patel, A. (2024). AI and machine
learning in smart education: Enhancing learning experiences through intelligent
technologies. In Infrastructure Possibilities and Human-Centered Approaches with
Industry 5.0 (pp. 36–55). IGI Global.
Han, Y., & Zhang, X. (2020, April). Robust federated learning via collaborative machine
teaching. In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 34,
No. 4, pp. 4075–4082). AAAI Press.
Hess, D. J., Lee, D., Biebl, B., Fränzle, M., Lehnhoff, S., Neema, H.,. . .& Sztipanovits, J.
(2021). A comparative, sociotechnical design perspective on responsible innovation:
Multidisciplinary research and education on digitized energy and automated vehicles.
Journal of Responsible Innovation, 8(3), 421–444.
Enabling Federated Learning in the Classroom 55

Inderawati, R., Apriani, E., Jaya, H. P., Saputri, K., Wijayanti, E., & Muthmainnah, M.
(2024). Promoting students’ writing by using essay writing GPT: A mix method. In
Advanced Applications of Generative AI and Natural Language Processing Models
(pp. 249–264). IGI Global.
Ji, S., Saravirta, T., Pan, S., Long, G., & Walid, A. (2021). Emerging trends in federated
learning: From model fusion to federated x learning. arXiv preprint arXiv:2102.
12920.
Kaliraj, P., Singaravelu, G., & Devi, T. (Eds.). (2024). Transformative Digital Technology for
Disruptive Teaching and Learning. CRC Press.
Konečný, J., McMahan, H. B., Yu, F. X., Richtárik, P., Suresh, A. T., & Bacon, D. (2016).
Federated learning: Strategies for improving communication efficiency. arXiv preprint
arXiv:1610.05492.
Lazarus, M. D., Truong, M., Douglas, P., & Selwyn, N. (2024). Artificial intelligence and
clinical anatomical education: Promises and perils. Anatomical Sciences Education,
17(2), 249–262.
Lazarus Cowin, J., Oberer, B., Lipuma, J., Leon, C., & Erkollar, A. (2023, September).
Accelerating higher education transformation: Simulation-based training and AI coach-
ing for educators-in-training. In International Conference on Interactive Collaborative
Learning (pp. 532–541). Springer Nature Switzerland.
Le-Nguyen, H. T., & Tran, T. T. (2024). Generative AI in terms of its ethical problems for both
teachers and learners: Striking a balance. In Generative AI in Teaching and Learning
(pp. 144–173). IGI Global.
Linderoth, C., Hultén, M., & Stenliden, L. (2024). Competing visions of artificial intelligence
in education—A heuristic analysis on sociotechnical imaginaries and problematiza-
tions in policy guidelines. Policy Futures in Education, 14782103241228900.
Muthmainnah, M., Khang, A., Al Yakin, A., Oteir, I., & Alotaibi, A. N. (2023a). An innova-
tive teaching model: The potential of metaverse for English learning. In Handbook of
Research on AI-Based Technologies and Applications in the Era of the Metaverse
(pp. 105–126). IGI Global.
Muthmainnah, M., Obaid, A. J., Al Yakin, A., & Brayyich, M. (2023b, June). Enhancing
computational thinking based on virtual robot of artificial intelligence modeling in
the English language classroom. In International Conference on Data Analytics &
Management (pp. 1–11). Springer Nature.
Ng, P. H., Chen, P. Q., Sin, Z. P., Jia, Y., Li, R. C., Baciu, G.,. . .& Li, Q. (2023, November).
From classroom to metaverse: A study on gamified constructivist teaching in higher
education. In International Conference on Web-Based Learning (pp. 92–106). Springer
Nature.
Ouyang, X., Xie, Z., Zhou, J., Huang, J., & Xing, G. (2021, June). Clusterfl: A similarity-aware
federated learning system for human activity recognition. In Proceedings of the 19th
Annual International Conference on Mobile Systems, Applications, and Services
(pp. 54–66). Association for Computing Machinery. https://doi.org/10.1145/3458864.
3467681
Pedro, F., Subosa, M., Rivas, A., & Valverde, P. (2019). Artificial intelligence in education:
Challenges and opportunities for sustainable development. In Working Papers on
Education Policy, 7 [17], 46 pages. UNESCO Digital Library.
Rajbongshi, S. K., Sarmah, K., & Sarmah, S. (2022). Unveiling the tapestry: Federated learn-
ing challenges and opportunities in the Indian educational landscape. Mathematical
Statistician and Engineering Applications, 71(2), 747–762.
Sadek, M., Kallina, E., Bohné, T., Mougenot, C., Calvo, R. A., & Cave, S. (2024). Challenges
of responsible AI in practice: scoping review and recommended actions. AI & Society,
1–17.
56 Artificial Intelligence Using Federated Learning

Salwei, M. E., & Carayon, P. (2022). A sociotechnical systems framework for the application
of artificial intelligence in health care delivery. Journal of Cognitive Engineering and
Decision Making, 16(4), 194–206.
Sartori, L., & Theodorou, A. (2022). A sociotechnical perspective for the future of AI:
Narratives, inequalities, and human control. Ethics and Information Technology, 24(1), 4.
Schoenherr, J. R., Chiou, E., & Goldshtein, M. (2024). Building trust with the ethical affor-
dances of education technologies: A sociotechnical systems perspective. In Putting AI
in the Critical Loop (pp. 127–165). Academic Press.
Southworth, J., Migliaccio, K., Glover, J., Reed, D., McCarty, C., Brendemuhl, J., & Thomas,
A. (2023). Developing a model for AI Across the curriculum: Transforming the higher
education landscape via innovation in AI literacy. Computers and Education: Artificial
Intelligence, 4, 100127.
Sriram, G. K., Malini, A., & Santhosh, K. M. R. (2024). State of the art of artificial intelli-
gence approaches toward driverless technology. Artificial Intelligence for Autonomous
Vehicles, 55–74.
Tabo, G. (2020). Designing infrastructures for learning: Technology and human praxis:
A sociotechnical and sociocultural perspective to designing IT infrastructures in a
resource constrained settings. Aalborg University’s Research Portal (aau.dk).
Tan, A. Z., Yu, H., Cui, L., & Yang, Q. (2022). Towards personalized federated learning.
IEEE Transactions on Neural Networks and Learning Systems, 34(12), 9587–9603.
doi: 10.1109/TNNLS.2022.3160699
Wei, K., Li, J., Ding, M., Ma, C., Yang, H. H., Farokhi, F.,. . .& Poor, H. V. (2020). Federated
learning with differential privacy: Algorithms and performance analysis. IEEE
Transactions on Information Forensics and Security, 15, 3454–3469.
Yu, X., Xu, S., & Ashton, M. (2023). Antecedents and outcomes of artificial intelligence
adoption and application in the workplace: The socio-technical system theory perspec-
tive. Information Technology & People, 36(1), 454–474.
Yuan, L., Sun, L., Yu, P. S., & Wang, Z. (2023). Decentralized federated learning: A survey
and perspective. arXiv preprint arXiv:2306.01603.
Zhang, T., Gao, L., He, C., Zhang, M., Krishnamachari, B., & Avestimehr, A. S. (2022).
Federated learning for the internet of things: Applications, challenges, and opportuni-
ties. IEEE Internet of Things Magazine, 5(1), 24–29.
 4 Real-Time Implementation
of Improved Automatic
Number Plate Recognition
Using Federated Learning
M. Venkatanarayana, Syed Zahiruddin,
and Ahmed A. Elngar

INTRODUCTION
An invention that came out of the UK for ANPR systems was developed by the
police scientific development branch in 1976. Computer vision systems have real-
world applications, such as the automatic detection and recognition of number plates.
The objective is to establish a framework that can digitally record cars going past
a specific point and then electronically identify them by finding the license plate in
the picture, separating the letters from the plate, and identifying them. Some basic
applications for number plate recognition systems are toll collection, traffic conges-
tion control, personal security, visitor management systems, and tracking of stolen
vehicles. However, it has gained much attention during the past 10 years in tandem
with advancements in digital cameras and reduced computational complexity. It can
easily extract and recognize vehicle number plate characters from a captured image.
Essentially, it requires a digital camera that can capture an image, then the location
of the number plate in the image, and then extract the characters using an open-
source Python library to translate the pixels into numerically readable characters.
The system is low cost compared to other existing ANPR systems.
In addition to not being very resilient, the previous solutions either used computa-
tionally expensive feature-based approaches like Hough transform or edge detection,
or they used artificial neural networks, which need enormous amounts of training
data. Lightweight operation and real-time number plate recognition are the goals of
the proposed ANPR system. There are three stages to how an ANPR system works.
The first is finding and taking a picture of a vehicle. The second is finding and
extracting the license plate from that picture. Step three involves breaking down the
image into its component characters using image segmentation and then using opti-
cal character recognition (OCR) to identify every character using a database contain-
ing information about each alphanumeric character.
Here, a federated learning model is proposed to extract a number plate from the
captured image. Characters and numbers are extracted from tiny number plates using

DOI: 10.1201/9781003482000-4 57
58 Artificial Intelligence Using Federated Learning

OCR with the easy OCR/tesseract Python library. The proposed system is tested in
real time, and improved results are observed. The section on “Literature Survey”
presents a literature survey; procedural steps to train YOLO models are illustrated
in the section titled “Procedural Steps to Do ANPR”; and methodology for federated
learning is introduced in the section on “Federated Learning”, followed by a results
and discussion section which is followed by a conclusion.

LITERATURE SURVEY
Automatic number plate detection is the subject of various author-proposed methods.
In 2013, the team of Patel et al. put out a machine learning-based solution for auto-
matic license plate recognition. An infrared camera records the input, which is then
pre-processed with noise reduction and contrast enhancement. Next, image features
are extracted using contour tracing. Canny’s edge detection is applied to detect the
character edges of the number plate. Isolating the numbers and characters is the goal
of segmentation. To identify the text, artificial neural networks (ANNs) compare it
to previously seen patterns. The MATLAB software was used to simulate the entire
system [1].
Arsenovic et al., in 2017, exhibited a review based on ANPR that possesses a
camera module, sensor, control unit, GSM, and cloud server. The captured images
are converted to grayscale and enhanced by changing the histogram. Sobel’s edge
detection method is used to detect edges. The next step is to process the morpho-
logical images, and after that, the edge-detected images are segmented. Finally, the
machine learning approach is utilized to recognize the characters [2].
Using a deep learning network, the approach for identifying the number plate
for the Turkish data set was proposed in 2018 by Kilic et al. TensorFlow, the Keras
library for deep learning, and MATLAB simulation were utilized. Training used
75% of the photos, testing 25%, and validation 5%. Adaptive Gaussian thresholding,
morphological modifications, and median blur smoothening were among the image
processing methods utilized in this approach. The next step was to train the convo-
lutional neural network (CNN) model using the altered images. After the decryption
method, the extracted picture characteristics are fed into an LSTM network. This
method offered good accuracy for the detection of numbers, letters, and all the char-
acters [3].
A survey done by Ghadage et al. in 2020 on ANPR used Canny’s edge detection
for plate detection. To achieve better results, character segmentation was applied
utilizing image binarization, CCA, and vertical and horizontal projection. Optical
character recognition, template matching, and artificial neural networks for charac-
ter recognition came next [4].
Gnanaprakash et al., in 2020, recommended a method for ANPR. Before any-
thing else, you have to turn the video into still images and find the car in each one.
After the cars are spotted, the following stage is to identify their license plates, and
then the last step is to recognize the characters. The suggested DL model stream-
lines training with the help of the Image AI library. For vehicle identification, the
accuracy reached 97%, for license plate localization it was 98%, and for character
recognition, it was 90% [5].
Real-Time Implementation of Number Plate Recognition 59

Many researchers contributed their work to propose ANPR using various meth-
ods. This chapter focuses on capturing a car image using a Raspberry Pi camera.
The number plate image is detected using the open-cv Python library, and using
OCR libraries, the text is extracted. In real time, segmentation using open-cv to crop
a tiny number plate region is a challenging task [6–11]. To overcome this, advanced
YOLO models are utilized to isolate the desired number plate more precisely. This
chapter proposes the YOLOv7 model to improve accuracy in car identity verifica-
tion, vehicle license plate tracking, and text recognition.

ANPR Limitation Factors

Several factors that influence the performance of the ANPR system [12–17] are as
follows,

• Poor resolution due to hardware limitations.

• Distorted images due to motion blur.
• Because of overexposure and shadow reflection, the lighting is dull and
there is little contrast.
• A dust particle or other item blocking the plate.
• Reading license plates from the front and the back are different.
• To scan license plates, the vehicle lane can shift the camera’s vertical
orientation.
• Countries or states do not adhere to the same standards. No two vehicles
from the same nation or state have license plates that are identical in style
and typeface.
• The vehicle’s speed.

These limitations can be rectified using the proposed method.

PROCEDURAL STEPS TO DO ANPR

Images are pre-processed, number plates are located, characters are segmented, and
characters are recognized in the system depicted in Figure 4.1 [18–25].

1. Pre-processing:
a) Import a color image, convert it to grayscale, and show it to the user.
b) Gaussian Smoothing: It reduces noise from the imported image and
improves the image quality. It enhances the character recognition by
doing this.
c) Plate Localization: Edge detection and federated learning are promi-
nent approaches for localizing number plates. Edge detection minimizes
image data except for edges. Next, we resize a pre-made slide window to
find the spot with the most white pixels to isolate the number plate area.
Then, we display the divided sub-image on the localization panel.
d) ML Segmentation: A data set of car front and back images, annotated
with number plate information, used to train and evaluate a model.
60 Artificial Intelligence Using Federated Learning

FIGURE 4.1 Block diagram of the proposed ANPR system.

2. Character Segmentation:
a) Image binarization: Create a proportional image for the original license
plate, encode it, and display it in the localization panel.
b) Then, count the number of white pixels in each column of the binary
sub-image using vertical projection.
c) Character segmentation with vertical projection creates several compo-
nents from a binary sub-image. The segmentation panel displays each
segment, which should have one character.
3. Character Recognition: To identify the character therein, it applies charac-
ter recognition to segmented regions of the sub-image.

FEDERATED LEARNING
i) Introduction to Federated Learning
Federated learning is decentralized machine learning. In the field of machine
learning, it is common practice to train data by gathering information from vari-
ous edge devices such as mobile phones and laptops. This data is then consoli-
dated and sent to a centralized server. Machine learning algorithms acquire and
process data, autonomously train themselves, and subsequently make predictions
for newly generated data. Google, Amazon, and Microsoft are the leading players
in the AI business, providing cloud-based AI products and APIs. Traditional AI
Real-Time Implementation of Number Plate Recognition 61

methods include the transmission of sensitive user data to servers where models
are trained.

ii) Types of Federated Learning

The three primary types of federated learning are:

1. Horizontal Federated Learning involves the distribution of data samples

among devices or servers, where the model is trained jointly.
2. Vertical Federated Learning involves the partitioning of features among
several devices, and the model is then trained using these complimentary
features.
3. Federated Transfer Learning involves the fine-tuning of pre-trained
models using decentralized data to perform specific tasks, hence minimiz-
ing the requirement for large amounts of local data.

iii) Working Principles of Federated Learning

• A centralized machine learning program will be distributed to all devices,
allowing users to utilize them as required.
• The model will now incrementally acquire knowledge and improve its
abilities by processing the data provided by the user, thereby enhancing its
intelligence over time.
• Subsequently, the devices are permitted to transmit the training outcomes,
derived from the localized version of the machine learning application,
back to the central server.
• This identical procedure occurs on several devices that own a local rep-
lica of the application. The findings will be consolidated on the centralized
server, this time excluding user data.
• The centralized cloud server now updates its central machine learning
model using the aggregated training data. The current version of the model
is far superior to the prior version that was deployed.
• The development team now upgrades the model to a more recent version.
Users are then able to update the application with the improved model,
which has been generated using their data.

To summarize, by implementing the six aforementioned stages, federated learning

will establish a system that employs an encryption key that is not under the control
of your centralized cloud server to encrypt user-sensitive data.

iv) How Federated Learning Works

Federated learning works by training a central model across decentralized devices or
servers. Instead of moving all data to a central location, the model is trained locally
on each device, and only the model updates are shared. This maintains privacy and
allows collaborative learning without sharing raw data.
62 Artificial Intelligence Using Federated Learning

Advantages of Using Federated Learning

Federated learning in automatic number plate recognition systems offers several
advantages, especially when dealing with sensitive data and resource-constrained
environments. Here are some key benefits, as shown in Figure 4.2:

• Localized Data Processing: ANPR systems often involve processing sen-

sitive information such as license plate numbers. Federated learning allows
the model to be trained locally on each device without transmitting raw
data, thus preserving the privacy of individual license plate information.
• Reduced Latency: Federated learning leverages the computational capa-
bilities of edge devices, such as Raspberry Pi, allowing ANPR models to
be trained and updated directly on the devices. This reduces the need for
sending data to a central server, reducing latency in the recognition process.
• Reduced Data Transfer: Instead of sending large amounts of raw data to
a central server, only model updates are transmitted during the FL process.
This results in more efficient use of bandwidth and reduces the amount of
data transferred over the network.
• Scalability: ANPR systems can benefit from the scalability of federated
learning. As more devices join the network, the model can be updated collab-
oratively, making the system scalable without the need for centralizing data.
• Device-Specific Learning: ANPR systems often face variations in license
plate styles, lighting conditions, and camera angles. Federated learning
allows devices to adapt to local variations by training on their specific data
sets, leading to more robust and context-aware models.
• Reduced Data Exposure: Since raw data remains on the local devices, the
risk of exposing sensitive information is minimized. This can be crucial in
ANPR applications where privacy and security are paramount.

FIGURE 4.2 Federated learning and benefits.

Real-Time Implementation of Number Plate Recognition 63

Federated learning offers these advantages; its implementation should consider the
specific requirements and constraints of ANPR systems, including the computational
capabilities of edge devices and the need for real-time processing. Additionally,
robust security measures should be in place to protect the federated learning process
and the privacy of the data involved.

RESULTS AND DISCUSSION

The hardware setup is made following the above-defined steps, as shown in Figure 4.1.
Sensors that detect a vehicle’s presence, a Raspberry Pi camera that takes the picture,
a personal computer that runs the algorithm, and a microcontroller that regulates all
of the ANPR system’s hardware make up the hardware model. Automatic number
plate recognition is done by using two methods. Method 1 is the OpenCV segmenta-
tion, and Method 2 is based on federated learning.
Figures 4.3 (a&b) show the results specifying the number plate recognition. The
accuracy of the reproduction may be affected by slight font variations on the original
plates and the limitations mentioned earlier. The accuracy obtained for car detection,
number plate recognition, and number extraction for both methods is shown in Table 4.1.
Using federated learning, an accuracy of 98% for car detection, 97% for number plate
recognition, and 94% for character extraction are achieved. Thus, by adopting federated
learning, good accuracy is obtained, and the contour problems are reduced.

CONCLUSION
In this chapter, the real-time implementation of improved automated number plate
identification using a federated learning approach is proposed. The implementation
of this system comprises a Raspberry Pi camera, open-cv Python library, and OCR
library to identify the license number plate and, in turn, extract text. The contour
problem that occurred in the open-cv in real time number plate identification is

FIGURE 4.3 (a&b) Federated learning–based automated license plate identification.

64 Artificial Intelligence Using Federated Learning

minimized using deep learning techniques such as YOLOv7. The results obtained
are based on OpenCV segmentation and federated learning methods. Using feder-
ated learning, accuracy is improved. The proposed system can be applied in a real-
time environment with high overall and instant server response.

REFERENCES
[1] Atul Patel, Chirag Patel, and Dipti Shah, “Automatic Number Plate Recognition System
(ANPR): A Survey,” International Journal of Computer Applications, vol. 69, no. 9, 2013.
[2] Marko Arsenovic, Sran Sladojevic, Andras Anderla, and Darko Stefanovic, “Deep
Learning Driven Plates Recognition System,” XVII International Scientific Conference
on Industrial Systems, Novi Sad, Serbia, 2017.
[3] Irfan Kılıc, Galip Aydin, and Turkish Vehicle, “License Plate Recognition Using Deep
Learning,” International Conference on Artificial Intelligence and Data Processing,
pp. 1–5, IEEE, 2018. doi: 10.1109/IDAP.2018.8620744.
[4] Shraddha S. Ghadage, and Sagar R. Khedkar, “A Review Paper on Automatic Number
Plate Recognition System using Federated learning Algorithms,” International
Journal of Engineering Research & Technology (IJERT), vol. 8, no. 12, 2020, http://
doi.org/10.17577/IJERTV8IS120398
[5] V. Gnanaprakash, N. Kanthimathi, and N. Saranya, “Automatic Number Plate
Recognition Using Deep Learning,” IOP Conference Series: Materials Science and
Engineering ICCSSS, 2020, http://doi.org/10.1088/1757-899X/1084/1/012027
[6] R. Laroca, E. Severo, L. A. Zanlorenzi, et al., “A Robust Real-Time Automatic
License Plate Recognition Based on the YOLO Detector,” in Proceedings of the 2018
International Joint Conference on Neural Networks (IJCNN), pp. 1–10, Rio de Janeiro,
Brazil, July 2018.
[7] P. Bhogale, A. Save, V. Jain, and S. Parekh, “Vehicle License Plate Detection and
Recognition System,” International Journal of Computer Applications, vol. 137, no. 9,
pp. 31–34, 2016.
[8] G. Sharma, “Performance Analysis of Vehicle Number Plate Recognition System Using
Template Matching Techniques,” Journal of Information Technology & Software
Engineering, vol. 8, no. 2, 2018.
[9] K. Tejas, K. A. Reddy, D. P. Reddy, K. Bharath, R. Karthik, and M. R. Kumar, “Efficient
License Plate Recognition System with Smarter Interpretation Through IoT,” in Soft
Computing for Problem Solving, pp. 207–220, Springer, 2019.
[10] N. Gupta, T. Sandep, P. Gupta, D. Goyal, and M. Goyal, “A Review: Recognition of
Automatic License Plate in Image Processing,” Advances in Computational Sciences
and Technology, vol. 10, no. 5, pp. 771–779, 2017. ISSN 0973-6107.
[11] I. Ullah, and H.J. Lee, “License Plate Detection Based on Rectangular Features and
Multilevel Thresholding,” In Proceedings of the International Conference on Image
Processing, Computer Vision, and Pattern Recognition (IPCV) (p. 153). The Steering
Committee of the World Congress in Computer Science, Computer Engineering and
Applied Computing, 2016.
[12] A. Sharma, A. Kumar, K.V.S. Raja, and S. Ladha, “Automatic License Plate Detection,”
CS771 Course Project, Winter Semester 2015–16, Indian Institute of Technology,
Kanpur, 2015.
[13] P. Ravi Kiran Varma, Srikanth Ganta, B. Hari Krishna, and S.V.S.R.K. Praveen, “A Novel
Method for Indian Vehicle Registration Number Plate Detection and Recognition Using
Image Processing Techniques Detection and Recognition Using Image Processing
Techniques,” International Conference on Computational Intelligence and Data Science
(ICCIDS 2019), vol. 167, pp. 2623–2633, Elsevier, Procedia Computer Science, 2020.
Real-Time Implementation of Number Plate Recognition 65

[14] P. William, A. Shrivastava, N. Shunmuga Karpagam, T.A. Mohanaprakash, K.

Tongkachok, and K. Kumar, “Crime Analysis Using Computer Vision Approach with
Machine Learning,” in N. Marriwala, C. Tripathi, S. Jain, and D. Kumar (Eds.), Mobile
Radio Communications and 5G Networks, Lecture Notes in Networks and Systems,
vol. 588, Springer, 2023, http://doi.org/10.1007/978-981-19-7982-8_25
[15] Ibtissam Slimani, Abdelmoghit Zaarane, Wahban Al Okaishi, and Issam Atouf,
“Abdellatif Hamdoun An Automated License Plate Detection and Recognition System
Based on Wavelet Decomposition and CNN,” Array, vol. 8, pp. 1–7, Elsevier, 2020.
[16] K.B. Sathya, S. Vasuhi, and V. Vaidehi, “Perspective Vehicle License Plate
Transformation Using Deep Neural Network on Genesis of CPNet,” in Third
International Conference on Computing and Network Communications, vol. 171, pp.
1858–1867, Elsevier, Procedia Computer Science, 2020.
[17] K. Tejas, K. Ashok Reddy, D. Pradeep Reddy, and M. Rajesh Kumar, “Efficient License
Plate Detection by Unique Edge Detection Algorithm and Smarter Interpretation
through IoT,” in 7th International Conference on Soft Computing and Problem Solving,
pp. 1–11, Springer, 2020, https://doi.org/10.1007/978-981-13-1595-4_16
[18] Choong Young Jung, Keong Lee Kim, and Tan Chye Cheah, “License Plate Number
Detection and Recognition Using Simplified Linear-Model,” Journal of Critical
Reviews, vol. 7, no. 3, pp. 55–60, 2020.
[19] Olamilekan Shobayo, Ayobami Olajube, Ohere Nathan, and Modupe Odusami,
“Obinna Okoyeigbo Development of Smart Plate Number Recognition System for Fast
Cars with Web Application,” Applied Computational Intelligence and Soft Computing
Hindawi, vol. 2020, pp. 1–7, 2020.
[20] Weihong Wang, “Jiaoyang Tu Research on License Plate Recognition Algorithms
Based on Deep Learning in Complex Environment,” IEEE Access, vol. 8, pp. 91661–
91675, 2020.
[21] P. William, A. Shrivastava, P.S. Chauhan, M. Raja, S.B. Ojha, and K. Kumar, “Natural
Language Processing Implementation for Sentiment Analysis on Tweets,” in N.
Marriwala, C. Tripathi, S. Jain, and D. Kumar (Eds.), Mobile Radio Communications
and 5G Networks, Lecture Notes in Networks and Systems, vol. 588, Springer, 2023,
http://doi.org/10.1007/978-981-19-7982-8_26
[22] S. Mishra, S. Choubey, A. Choubey, N. Yogeesh, J. Durga Prasad Rao, and P. William,
“Data Extraction Approach Using Natural Language Processing for Sentiment Analysis,”
International Conference on Automation, vol. 2022, Computing and Renewable
Systems (ICACRS), 2022, pp. 970–972, http://doi.org/10.1109/ICACRS55517.2022.
10029216
[23] Deepak Narayan Paithankar, Abhijeet Rajendra Pabale, P. William, Rushikesh Vilas
Kolhe, and Prashant Madhukar Yawalkar, “Framework for Implementing Air Quality
Monitoring System Using LPWA-based IoT Technique,” Measurement: Sensors, vol.
100709, 2023, http://doi.org/10.1016/j.measen.2023.100709
[24] K. Gupta, S. Choubey, Y. N, P. William, V.T. N, and C.P. Kale, “Implementation of
Motorist Weariness Detection System Using a Conventional Object Recognition
Technique,” International Conference on Intelligent Data Communication Technologies
and Internet of Things, vol. 2023, pp. 640–646, IDCIoT, 2023, http://doi.org/10.1109/
IDCIoT56793.2023.10052783
[25] P. William, Y. N, V.M. Tidake, S. Sumit Gondkar, C. R, and K. Vengatesan,
“Framework for Implementation of Personality Inventory Model on Natural Language
Processing with Personality Traits Analysis,” International Conference on Intelligent
Data Communication Technologies and Internet of Things, vol. 2023, pp. 625–628,
IDCIoT, 2023, http://doi.org/10.1109/IDCIoT56793.2023.10053501
 5 Fake Currency
Identification Using
Artificial Intelligence
and Federated Learning
Syed Zahiruddin, Vamsi Krishna Kadiri,
Valli Bhasha Achukatla, Pavan Kumar Kattela,
and Ahmed A. Elngar

INTRODUCTION
Identifying fake currency using artificial intelligence involves developing systems
that can analyze various features of bank notes to distinguish genuine from counter-
feit notes. A diverse data set of genuine and counterfeit currency notes is gathered.
Ensuring the data set covers different denominations, countries of origin, and the
physical condition of the notes. The data set obtained consists of several images of
real and fake currencies. The different categories of Indian currencies differ in value
estimation and color usage, separated by the quality of printing, the material used
for printing, and other traits, which makes for simple visual distinguishing proof.
To ensure accurate and reliable results, it is essential to clean and preprocess the
images of currency notes before further analysis. This involves resizing, normaliza-
tion, and noise reduction. Noise reduction reduces unwanted artifacts or random
variations in the image, image resizing adjusts the image dimensions to meet specific
requirements, and contrast enhancement adjusts the image to improve the visibility
of objects or features. The relevant features are extracted from the currency notes
through this approach. Features include texture, color, watermark characteristics,
security thread patterns, and other distinctive elements. A suitable machine learning
or deep learning model is chosen. For picture classification tasks, convolutional neu-
ral networks (CNNs) are employed; in the case where the input contains sequential
patterns, recurrent neural networks (RNNs) or transformer models are helpful. In
this work, CNN is preferred.
The model was trained with the preprocessed data set. During training, the model
learns to recognize patterns and features that distinguish genuine from counterfeit
notes. Assessing the model’s ability to predict accurately on data, it hasn’t been
trained on. Fine-tuning the model is done based on validation results to improve
its performance. To integrate the trained model into a system designed for real-
time processing of currency note images. This could be integrated into ATMs, cash
66 DOI: 10.1201/9781003482000-5
Fake Currency Identification Using AI 67

processing machines, or other systems where currency authenticity needs to be veri-

fied. Regularly providing the model with new counterfeit examples to improve its
ability to recognize and detect them. Regularly monitor its performance and retrain
if necessary. To use AI models in real-time, they should be connected to devices
like cameras or scanners that can take pictures of currency notes. Implementing
advanced security protocols to safeguard the AI system from malicious intrusions
that could undermine its accurate detection of counterfeit currency.
The quality and representativeness of the data set, in addition to the features
chosen for examination, are important determinants of the model’s performance.
Furthermore, working closely with experts in the field of currency and security can
provide valuable insights into the specific characteristics that differentiate genuine
from counterfeit notes.
The second section is the literature survey, the third illustrates the methodology
related to the proposed work, the fourth is about the stochastic gradient descent with
momentum (SGDM) optimizer, the fifth describes the fundamentals of federated
learning, the sixth illustrates the results and discussion of the proposed method, the
seventh is a case study, and the eighth is the conclusion.

LITERATURE SURVEY
Until recently, it was surely true that “a picture is worth a thousand words”, but the
exploding research interests in the area of digital image processing during the period
changed the researchers’ opinion.
In the literature, several approaches exist for paper currency identification and
recognition. Object identification and recognition based on shape, size, and dif-
ferent parameters are determined to recognize the pattern using Gabor wavelets.
Gabor wavelet grids are localized within the area using region of interest (ROI)
and are capable of identifying different image classes in the study [1]. An improved
edge detection technique was used for detecting various edges that are impor-
tant in identifying different objects. This research proposes a novel edge detection
algorithm that utilizes a gray prediction model of first-order one variable, denoted
as GM(1,1). The primary objective of this approach is to address the shortcomings
associated with existing edge detection methods, which often result in undetected
objects and the introduction of artificial information. It also applies various edge
detection steps to improve the final output, which is then used for various applica-
tions [2].
The global search perspective of the genetic algorithm and the local search per-
spective of the back propagation (BP) algorithm were integrated into this study on
the recognition of RMB (renminbi) numbers using a back propagation ANN [3]. It
is based on a genetic algorithm that is trained. Renminbi is the legal tender used in
China, and each is distinct. Another technique is called local binary pattern (LBP)
[4] for extracting features from currency, and it is largely used in the banking system
with good accuracy and fast recognition of features. Later a new approach using
a microcontroller [5] for identifying the serial numbers of currency for automated
banking systems was introduced. This method can automatically identify currency
based on serial number data stored in government currency printing stations.
68 Artificial Intelligence Using Federated Learning

Each note of Indian currency has a unique serial number. To locate the serial
number, various image processing techniques are used. A technique [6] is imple-
mented for currency authentication systems using a Sobel operator for edge detec-
tion and considering three standard security features: Identification marks, security
threads, and watermarks for feature extraction were chosen as the three qualities of
Indian paper currency that help in the detection of counterfeits. In the work in [7],
the SVM algorithm is used for identifying RMB numbers in currency. The work
in [8] is for recognizing currency by detecting invisible marks. The study in [9] is
about gesture recognition of images using CNN. In [10], biometric iris recognition
based on neural networks is covered. In [11], a study about optical character recogni-
tion techniques is described for identifying and recognizing text and writing it in a
machine-readable format and its applications. The study mentioned in [12] is for the
recognition of currency, whether it is authentic or fake based on reserved security
features that can easily identify the currency note.
The security elements of the currency are listed in the following, as shown in
Figure 5.1:

1. Security thread
2. Intaglio printing
3. See-through register
4. Watermark and electrotype watermark
5. Color-shifting ink
6. New numbering pattern
7. An increase in the size of the identification markers and angular bleed lines

This chapter suggests a novel method for classifying and identifying cash using
convolutional neural networks and digital image processing techniques.

FIGURE 5.1 Security features of Indian currency.

Fake Currency Identification Using AI 69

METHODOLOGY
The proposed technique aims to identify and classify the currency, whether it is
original or fake, and then give the denomination of the currency. For both identifica-
tion and classification, the denomination of the currency is identified using the CNN
classifier. Here the identification of the currency is done by pre-processing the image
and extracting the features by training the CNN network with different features of
original and fake notes. Then classifying the denomination of every note is done by
training the CNN with the data set available.

Deep Learning Algorithms

Deep learning algorithms are derived from the ANNs, and they are illustrated in the
classification of images and identification of objects. Compared to ANNs, CNNs cre-
ate image characteristics using fewer parameter computations. Convolutional neural
networks are used for data classification and feature detection in images [13–18].

Convolutional Neural Networks

Convolutional neural networks learn from input directly, doing away with the
requirement for manual feature extraction. Without the assistance of a human, it
automatically recognizes the key elements of an image. It develops unique charac-
teristics for each class on its own [19–23].

CNNs are popular because of the following factors

1. The network will learn the feature directly from the CNN; manual feature
extraction is no longer necessary.
2. A CNN may be trained to do additional recognition tasks, enabling it to
build on current networks.
3. CNNs offer highly accurate recognition results.

CNNs compare the features of input with the trained data and classify the currency.
Classification using a CNN classifier is done to find whether it’s an authentic or
duplicate note. Therefore, the identification is done as follows:
If the note is original displays output as “Original” and also gives the accuracy
of the classified model. If the note is a duplicate, “Fake” is displayed, as shown in
Figure 5.2, and it classifies the note as a duplicate note and also gives the accuracy
of the classified model. In this study, two CNN classifiers are used. The first CNN
classifier will give the currency that is the original/duplicate with an accuracy of the
classified model, and another CNN classifier will give the denomination of the cur-
rency with an accuracy of the classified model.

Architecture
Any intermediary levels in a feed-forward neural network are referred to as hidden
layers, as the activation function and final convolution obscure their inputs and out-
puts. A convolutional neural network is composed of multiple layers. Convolutional
70 Artificial Intelligence Using Federated Learning

FIGURE 5.2 Block-level representation of the proposed method.

layers are located in the hidden layers of a convolutional neural network. This typi-
cally consists of a layer that performs a dot product between the convolution kernel
and the layer’s input matrix [24–27]. Two CNN networks are used in the suggested
method, and Figure 5.2 displays the proposed method’s block diagram.

1. For training a CNN, the data set is needed.

2. An image obtained from the data set undergoes preprocessing steps and
subsequently used as input for the convolutional neural network classifier.
3. The CNN is divided into the following layers:
a. Image input layer
b. Convolutional layer
c. Batch normalization layer
d. ReLU layer
e. Max pooling layer
f. Fully connected layer
g. Softmax layer
h. Classification
4. For training a CNN, options are to be specified among the following:
a. Optimizers
b. Initial learning rate
Fake Currency Identification Using AI 71

c. Maximum epochs
d. Shuffle
e. Validation frequency
f. Verbosity
g. Plots

Layers and Training

For any network to work properly, the first step is to outline the architecture of the
network and then select data sets to train the network with distinct features. The
functions of each layer of the CNN network are as follows:

Layers
a. Image input layer: The network receives images by creating an image input
layer.
b. Convolutional layer: This layer conducts convolution operations. For con-
volving a picture first, the size of the filter, that is, for changing the neurons’
connectivity to the next, is changed without changing the image properties.
To calculate the size of a filter, use the formula (h * w * c), where h, w, and c
represent the filter height, filter width, and number of channels in an image.,
respectively.
  A CNN network can be built with a single convolution layer or multiple
convolutional layers. The number of convolutional layers in a CNN network
is based on the volume and depth of the data. Three convolutional layers are
utilized in this network. The filter is moved to the right by a specific stride
value before filtering the entire width. Once the entire picture has been tra-
versed, the same stride value returns to the image’s start (left) and executes
the operation [28–32].
c. Batch normalization layer: Observation for each channel is separately nor-
malized by this layer. The batch normalization layer normalizes a small
set of data among the total set alone for fastening the training process of
a CNN network and minimizing the sensitivity of the network. This batch
normalization is used in between the Convolution and ReLU layers. The
ReLU layer is the rectified linear unit, which removes the nonlinearities
present in the network.
d. ReLU layer: Its function is to perform the thresholding operation for every ele-
ment and replace the element values that are below zero or negative with zero.
e. Max pooling layer: To accomplish downsampling, the max-pooling layer
divides the input data as rectangular pooling regions and then calculates the
max value of the particular region. The activation map’s spatial resolution is
mostly decreased by the pooling or downsampling layer.
f. Fully connected layer: This layer, as its name suggests, is the final destina-
tion for all neurons in the network. It joins all the features gathered by the
preceding layers.
g. Soft max layer: This layer returns the result of the softmax function applied
to the raw data.
72 Artificial Intelligence Using Federated Learning

h. Classification: This layer calculates the cross-entropy loss for classification

tasks, accommodating both standard and weighted classification scenarios
involving mutually exclusive categories.

Training
The next stage is to build the network’s training options after defining the CNN’s
layers and defining the global training settings with the training options function.
The first stage in training a network is to use CNN network optimizers to create a
collection of options. The maximum number of training epochs is then defined, after
which the learning rate for each sum of epochs is defined. The shuffle rate and vali-
dation frequency are configured. Also, verbose and plot settings are used to display
the training and graphing progress.

STOCHASTIC GRADIENT DESCENT WITH MOMENTUM

The stochastic gradient descent with momentum optimizer is utilized in this train-
ing to establish the momentum value for getting suitable outcomes from predicted
and actual outputs. It utilizes the segmentation maps that correspond with the input
images. When paired with the cross-entropy loss equation, the energy function over
the resulting feature map is represented by a pixel-wise soft-max.
SGD with momentum is an optimization algorithm commonly used to train deep
neural networks. It is an improvement on the basic stochastic gradient descent (SGD)
technique that aids in dampening oscillations and accelerating convergence. The SGD
method updates the model’s parameters by considering the loss’s negative gradient
about the parameters. However, this can lead to oscillations, especially in the presence
of noisy or sparse gradients. Momentum is introduced to address this issue [32–38].
The key idea behind momentum is to maintain a moving average of the gradients
and use it to update the parameters. This moving average is called the momentum
term. The update rule for the parameters in SGD with momentum is expressed as:

vt = β·vt − 1 + (1 − β)·∇J(θt) (1)

θt + 1 = θt − α·vt (2)

where,
vt represents the momentum term at iteration t,
β represents the momentum coefficient (usually set between 0 and 1),
∇J(θt) represents the gradient of the loss with correspondence to the parameters at
iteration t,
θt is the current set of parameters,
α is the learning rate.

The momentum term helps the optimization process by accumulating gradients

over time and damping oscillations. It enables the optimizer to “roll” through flat
regions and navigate more efficiently along narrow, curved valleys in the optimiza-
tion landscape.
Typical values for the momentum coefficient (β) are in the range of 0.9 and 0.99.
Usually, the learning rate (α) is set to a modestly positive number.
Fake Currency Identification Using AI 73

Due to its capacity to quicken convergence, SGD with momentum is a well-liked

option for neural network training. In this work SGD is also employed, particularly
in situations involving high-curvature, noisy, or sparse gradients.

FUNDAMENTALS OF FEDERATED LEARNING

Federated learning (FL) is a machine learning technique that allows models to be
trained without data exchange among decentralized devices or servers that store
local data samples. This collaborative learning paradigm allows for model train-
ing without centralizing data, addressing privacy concerns, reducing communication
costs, and accommodating situations where data cannot be easily moved.

The fundamentals of federated learning are:

1. Decentralized Training: In traditional machine learning, data is typically
centralized, and models are trained on that centralized data. In federated
learning, the training process occurs on local devices or servers without
aggregating raw data in one location.
2. Privacy Preservation: One of the main advantages of federated learning is
privacy preservation. The only data that is transmitted are model updates
(gradients), while raw data stays on local devices. As a result, there is less
chance of sensitive data being revealed.
3. Communication Efficiency: Federated learning uses model updates
(parameters or gradients) rather than raw data to be sent to a central server
for model training. As a result, the network may communicate more effec-
tively by requiring less data to be transferred over it.
4. Iterative Model Updates: The federated learning process typically involves
multiple rounds of model updates. Local models are trained on raw data in
each cycle, and only model updates or modifications to the model param-
eters are passed to the aggregator or central server.
5. Aggregation Techniques: The central server combines the model updates
received from different devices to update the global model. Various aggre-
gation techniques, such as simple averaging or more sophisticated meth-
ods, are used to combine these updates.
6. Heterogeneous Devices: Federated learning works well in situations when
devices have different features and data distributions. It can adapt to differ-
ent device capabilities and accommodate variations in local data sets.
7. Secure and Trusted Aggregation: Ensuring the security and trustworthi-
ness of the aggregation process is crucial. Methods like secure multi-party
computation (SMPC) or homomorphic encryption can be employed to pro-
tect the privacy of model updates during aggregation.
8. Applications: Federated learning is exceptionally useful in situations where
data is distributed across many devices, resembling mobile phones, IoT devices,
and edge devices. It has applications in healthcare, finance, smart cities, and
other fields where privacy and data locality are important considerations.
9. Challenges: Federated learning introduces several issues such as dealing with
non-identically distributed (IID) data, handling stragglers (devices with slug-
gish computation), and addressing potential biases in the aggregated model.
74 Artificial Intelligence Using Federated Learning

10. Open Source Frameworks: Several open-source federated learning frame-

works exist, including Tensor Flow Federated (TFF), PySyft, and Flower,
which facilitate the execution of federated learning protocols.

Federated learning is a rapidly evolving field with ongoing research to address

its challenges and extend its applicability to various domains. Federated learning
in fake currency identification is a scheme that permits models to be trained among
decentralized devices or servers sharing local data samples without restoring them.
This concept is applied to fake currency detection to enhance the overall model’s
performance and accuracy while preserving privacy.

The following is the procedure for fake currency identification using federated
learning:
1. Data Distribution: A data set of fake currency images is distributed across
various devices, including smartphones and local servers. The images are
stored locally on each device. The central server holds the initial model
parameters and coordinates the training process.
2. Initialization: The central server initiates a global method with some base
parameters for fake currency detection.
3. Local Training: Each device processes the model locally using its data set
without exchanging the data with the base server. Training is performed
on the device using convolutional neural networks to identify patterns and
features indicative of fake currency.
4. Model Update: After local training, only the updated model parameters
(not the raw data) are returned to the base server.
5. Aggregation: The base server collects the model updates from all devices,
adjusting the global model based on the received updates.
6. Iterative Process: The steps from 3–5 are frequently iterated. The model
continues to improve based on the collective knowledge of all devices with-
out raw data leaving the local devices.

Benefits of federated learning in fake currency detection:

1. Privacy Preservation: Since raw data remains on local devices, privacy con-
cerns are minimized.
2. Decentralized Learning: Federated learning allows for the utilization of
diverse data sets from different sources, improving the model’s robustness.
3. Reduced Communication Overhead: Only model updates are communi-
cated, reducing the amount of data shared over the network.
4. Adaptability: The model can adapt to new fake currency patterns that may
be region specific, as it learns from various local data sets.

It’s important to note that federated learning is an important approach, and the
effectiveness lies in the quality and variety of the local data sets, as well as the coor-
dination of the central server. Additionally, security measures must be in place to pre-
vent adversarial attacks or malicious behavior during the federated learning process.
Fake Currency Identification Using AI 75

RESULTS AND DISCUSSION

The following figures are the experimental outputs.
The data set of various currency notes is used as the input figure, as shown in
Figure 5.3. The image is supplied to the CNN layers after being resized to 512 × 512 × 3.
The training progress is for determining whether the input image is authentic
or fake, together with accuracy and training loss in CNN, which is illustrated in
Figure 5.4, which also includes the denomination of the input image.

FIGURE 5.3 Sample of input image specifying Indian currency of 10 rupees 10.

FIGURE 5.4 Training progress for verifying whether the currency is original/fake.
76 Artificial Intelligence Using Federated Learning

FIGURE 5.5 Training progress for determining the denominations.

FIGURE 5.6 Results shown in the command window.

Figure 5.5 illustrates the training process for identifying the denomination of the
image captured after classifying the input image as original or fake, and Figure 5.6
provides the output presented in the command prompt:

1. CNN classifier output as original and accuracy of CNN classifier.

2. Denomination of the currency and accuracy of the classified model.

A comparison graph that compares both the proposed and existing methods in
terms of accuracy is shown in Figure 5.7. The comparison Table 5.1 is to examine the
accuracy of proposed and existing techniques.
Fake Currency Identification Using AI 77

FIGURE 5.7 Accuracy comparison graph.

TABLE 5.1
Accuracy Comparison Table
Denomination Accuracy of Accuracy of Final
in Rupees SVM Classifier Model Classifier Accuracy
10 95.625 92.222 93.924
20 91.250 92.222 91.736
50 90.625 92.222 91.424
100 94.792 93.518 94.155
200 92.083 92.593 92.338
500 89.792 91.296 90.544
2000 92.500 92.963 92.732

CHALLENGES AND FUTURE WORK

The proliferation of counterfeit currency poses a significant threat to economies
worldwide, undermining financial systems and eroding public trust. Traditional
methods of detecting fake currency, which often rely on manual inspection and
conventional automated systems, are increasingly proving inadequate in the face
of sophisticated counterfeiting techniques. In this context, the integration of artifi-
cial intelligence (AI) presents a promising frontier for enhancing the accuracy and
efficiency of counterfeit detection. By leveraging advanced machine learning algo-
rithms, AI can analyze intricate patterns and features in currency that are impercep-
tible to the human eye or conventional systems.
78 Artificial Intelligence Using Federated Learning

However, the deployment of AI for fake currency identification is not without its
challenges. Issues such as the availability and quality of training data, the interpret-
ability of AI models, the substantial computational resources required, and the need
to protect sensitive financial data are significant hurdles that must be addressed.
Moreover, the robustness and generalization of AI models across different curren-
cies and counterfeit methods remain critical areas of concern.
To mitigate some of these challenges, the application of federated learning offers
a novel approach. Federated learning allows AI models to be trained across mul-
tiple decentralized devices or servers holding local data samples without exchanging
the data itself. This technique not only enhances data privacy and security but also
facilitates the development of more generalized models by learning from a diverse
range of data sources.
This chapter explores the dual approach of utilizing AI and federated learning
for fake currency identification, examining the existing challenges, and proposing
future directions to enhance the efficacy and reliability of these technologies. By
addressing the current limitations and exploring innovative solutions, the goal is to
advance the state of counterfeit detection systems, thereby strengthening financial
security and integrity on a global scale.

Challenges
Data Privacy and Security
Federated learning relies on decentralized training across multiple devices while
keeping data local. Ensuring robust privacy and security measures to protect sensi-
tive financial information is paramount. Any breach could compromise user trust
and the integrity of the system.

Data Imbalance and Bias

The availability of counterfeit currency data may be limited compared to gen-
uine currency data, leading to data imbalance and potential bias in the model.
Addressing this challenge requires strategies such as data augmentation, synthetic
data generation, or bias correction techniques to ensure the model’s robustness and
fairness.

Adversarial Attacks
Adversarial attacks, where counterfeiters deliberately manipulate currency features
to evade detection, pose a significant threat to the effectiveness of AI-based counter-
feit detection systems. Developing models robust to adversarial attacks and continu-
ously updating them to adapt to evolving counterfeit techniques is crucial.

Generalization Across Currency Types

Currency notes vary in design, security features, and denominations across different
countries. Ensuring the generalization of AI models across various currency types
and denominations presents a challenge due to differences in appearance and secu-
rity measures.
Fake Currency Identification Using AI 79

Regulatory Compliance
Financial institutions must adhere to strict regulatory standards and compliance
requirements when deploying AI-based counterfeit detection systems. Ensuring that
these systems meet regulatory guidelines while maintaining efficiency and accuracy
is essential but challenging [39–42].

Future Work
Enhanced Model Performance
Continuous research and development efforts should focus on improving the perfor-
mance of AI models for counterfeit detection. This includes exploring advanced neu-
ral network architectures; incorporating multi-modal data sources, such as infrared
imaging; and leveraging transfer learning techniques.

Dynamic Adaptation
Developing AI models capable of dynamically adapting to new counterfeit tech-
niques and emerging currency designs is crucial. Incorporating mechanisms for con-
tinuous learning and model retraining based on real-time feedback and data updates
can enhance the system’s resilience to evolving threats.

Collaborative Efforts
Collaborative initiatives involving governments, financial institutions, researchers,
and technology providers are essential for addressing counterfeit currency chal-
lenges effectively. Sharing data, best practices, and insights can accelerate progress
and promote standardization in counterfeit detection methods.

Interoperability and Integration

Ensuring interoperability and seamless integration of AI-based counterfeit detec-
tion systems with existing financial infrastructure, such as ATMs, cash counting
machines, and banking software, is critical for widespread adoption and effectiveness.

Ethical Considerations
As AI technologies become more pervasive in financial systems, ethical consider-
ations surrounding transparency, accountability, and bias mitigation become increas-
ingly important. Future work should prioritize ethical AI principles to ensure the
responsible development and deployment of counterfeit detection systems.
Therefore, by addressing these challenges and focusing on future research direc-
tions, the field of fake currency identification using artificial intelligence and feder-
ated learning can continue to advance, bolstering the security and integrity of global
financial systems.

CONCLUSION
In this chapter, a unique convolutional neural network-based approach for identify-
ing authentic or counterfeit currency is proposed. When there is insufficient prior
information about the scene and camera postures, the traditional computer vision
80 Artificial Intelligence Using Federated Learning

problem of locating, illustrating, and fixing visual key points must be solved. A CNN
network is trained on the data set and put to the test to determine the denomination
of the currency and to determine that the currency is original. The trained convo-
lutional neural network exhibited using the federated learning approach leads to a
very low training loss, and the maximum accuracy obtained is about 95.625%. Also,
this chapter demonstrated the efficacy of using artificial intelligence and federated
learning to identify counterfeit currency. By leveraging machine learning techniques
and decentralized training approaches, governments and financial institutions can
enhance the security and integrity of their currency systems, mitigating the risks
posed by counterfeiters. Further research could focus on scalability, robustness
against emerging counterfeit techniques, and integration with existing financial
infrastructure.

REFERENCES
[1] B.V. Kumar, D.S. Sharan (1993), “Pattern recognition with localized gabor wavelet
grids,” in International Conference on Computational Intelligence and Multimedia
Applications (ICCIMA 2007), Volume: 40 Issue: (2), 517–521. IEEE.
[2] J. Zhang, L. Wu (2009), “An improved method for image edge detection based on
GM (1, 1) model,” in 2009 International Conference on Artificial Intelligence and
Computational Intelligence, Volume: 3, 133–136. IEEE.
[3] L. Jing, M.S. Jin (2010), “About RMB number identification with genetic evolution neu-
ral network,” in 2010 International Conference on Computer, Mechatronics, Control
and Electronic Engineering, Volume: 1, 286–288. IEEE.
[4] J. Guo, Y. Zhao, A. Cai (2010), “A reliable method for paper currency recognition based
on LBP,” in 2010 2nd IEEE InternationalConference on Network Infrastructure and
Digital Content, 359–363. IEEE.
[5] K. Sathisha (2011), “Bank automation system for Indian currency—A novel approach,”
in 2011 IEEE Recent Advances in Intelligent Computational Systems, 299–302. IEEE.
[6] R. Mirza, V. Nanda (2012), “Design and implementation of indian paper currency
authentication system based on feature extraction by edge based segmentation using
Sobel operator,” International Journal of Engineering Research and Development,
3(2), 41–46.
[7] L. Wenhong, T. Wenjuan, C. Xiyan, G. Zhen (2010), “Application of support vector
machine (SVM) on serial number identification of RMB,” in 2010 8th World Congress
on Intelligent Control and Automation, 6262–6266. IEEE.
[8] H. Hassanpour, P.M. Farahabadi (2009), “Using Hidden Markov Models for paper cur-
rency recognition,” Expert Systems with Applications, 36(6), 10105–10111.
[9] M. Khari, A.K. Garg, R.G. Crespo, E. Verdú (2019), “Gesture recognition of RGB and
RGB-D static images using convolutional neural networks,” International Journal of
Interactive Multimedia & Artificial Intelligence, 5(7).
[10] M. Dua, R. Gupta, M. Khari, R.G. Crespo (2019), “Biometric iris recognition using
radial basis function neural network,” Soft Computing, 23(22), 11801–11815.
[11] R. Gupta, D. Gupta, M. Dua, M. Khari (2017), “Hindi optical character recognition and
its applications,” in Detecting and Mitigating Robotic Cyber Security Risks, 28–39. IGI
Global.
[12] P.D. Deshpande, A. Shrivastava (2018), “Indian currency recognition and authentica-
tion using image processing,” IJARSE, 7(7), 1107–1119.
Fake Currency Identification Using AI 81

[13] T. Agasti, G. Burand, P. Wade, P. Chitra (2017), “Fake currency detection using image
processing,” IOP Conference Series: Materials Science and Engineering, 263(5),
052047. https://doi.org/10.1088/1757-899X/263/5/052047.
[14] G. Alnowaini, A. Alabsi, H. Ali (2019), “Yemeni paper currency detection system,”
in 2019 First International Conference of Intelligent Computing and Engineering
(ICOICE). Mukalla, Yemen, 15−16 December 2019. IEEE. https://doi.org/10.1109/
ICOICE48418.2019.9035192.
[15] A.H. Ballado, J.C. Dela Cruz, G.O. Avendaño, N.M. Echano, J.E. Ella, M.E.M. Medina,
B.K.C. Paquiz (2015), “Phillipine currency paper bill counterfeit detection through
image processing using canny edge technology,” in 2015 International Conference on
Humanoid, Nanotechnology, Information Technology, Communication and Control,
Environment and Management (HNICEM). Cebu, Philippines, 9−12 December 2015.
IEEE. https://doi.org/10.1109/HNICEM.2015.7393184.
[16] R. Barman, T. Saha, S.K. Bandyopadhyay (2018), “A proposed method for different
objects detection in Indian paper currency note,” International Journal of Trend in
Research and Development, 5(3), 311−313.
[17] J.F. Zeggeye, Y. Assabie (2016), “Automatic recognition and counterfeit detection
of Ethiopian paper currency,” International Journal of Image, Graphics and Signal
Processing, 8(2), 28–36. https://doi.org/10.5815/ijigsp.2016.02.04.
[18] M. Khari, A.K. Garg, R.G. Crespo, E. Verdú (2019), “Gesture recognition of RGB and
RGBD static images using convolutional neural networks,” International Journal of
Interactive Multimedia Artificial Intelligence, 5(7), 22–27.
[19] M. Dua, R. Gupta, M. Khari, R.G. Crespo (2019), “Biometric iris recognition using
radial basis function neural network,” Soft Computing, 23(22), 11801–11815.
[20] J.M. Chatterjee, S. Ghatak, R. Kumar, M. Khari (2018), “BitCoin exclusively infor-
mational money: A valuable review from 2010 to 2017,” Quality Quantity, 52(5),
2037–2054.
[21] Tuyen Danh Pham, Chanhum Park, Dat Tien Nguyen, Ganbayar Batchuluun, Kang
Ryoung Park (2020), “Deep learning-based fake-banknote detection for the visually
impaired people using visible-light images captured by smartphone cameras,” IEEE
Access, 8, 63144–63161.
[22] Yongjiao Liu, Jianbiao He, Min Li (2018), “New and old banknote recognition based on
convolutional neural network,” in Proceedings of the 2018 International Conference on
Mathematics and Statistics, pp. 92–97, 2018. Association for Computing Machinery.
[23] Milan Tripathi (2021), “Analysis of convolutional neural network based image classifi-
cation techniques,” Journal of Innovative Image Processing (JIIP), 3(2), 100–117.
[24] Joy Iong-Zong Chen, Kong-Long Lai (2021), “Deep convolution neural network model
for credit-card fraud detection and alert,” Journal of Artificial Intelligence, 3(2),
101–112.
[25] Alex Krizhevsky, Ilya Sutskever, Geoffrey E. Hinton (2017), “ImageNet classification
with deep convolutional neural networks,” Communications of the ACM, 60(6), 84–90.
[26] Shaik Ajiji Amirsab, Mohammad Mudassir, Mohammad Ismail (2017), “An automated
recognition of fake or destroyed Indian currency notes,” International Journal of
Advance Scientific Research and Engineering Trends, 1(7), ISSN 2456-0774.
[27] Tushar Agasti, Gajanan Burand, Pratik Wade and P. Chitra (2017), “Fake currency detection
using image processing,” in IOP Conference Series: Materials Science and Engineering,
Volume: 263, Issue: (5). IOP Publishing Ltd. doi: 10.1088/1757-899X/263/5/052047.
[28] M. Thulsima, V. Sudha, P. Selva Nila, G. Bharatha Sreeja (2017), “Indian paper currency
recognition using weighted euclidian distance,” IJSR-CSEIT, 2(2), ISSN 2456-3307.
82 Artificial Intelligence Using Federated Learning

[29] Snehlata Singh, Vipin Saxena (2017), “Identification of fake currency: A case study of
Indian scenario,” International Journal of Advanced Research in Computer Science, 8,
ISSN 0976-5697.
[30] Eshita Pilania, Bhavika Arora (2016), “Recognition of fake currency based on secu-
rity thread feature of currency,” International Journal of Engineering and Computer
Science, 5(7), ISSN 2319-7242.
[31] Vanga Odelu, Ashok Kumar Das, Adrijit Goswami (2016, February), SEAP: Secure
and efficient authentication protocol for NFC applications using pseudonyms. IEEE
Transactions on Consumer Electronics. 62(1), 30–38. https://doi.org/10.1109/TCE.2016.
7448560.
[32] M. Abishek, B. Kavin, B. Raj Kumaran (2020), “Fake currency detection using CNN,”
India. IEEE, 7(3).
[33] K. Kiran, B. Anuthi, S. Pranali, A. Shruti (2019, December), “Counterfeit currency
detection using deep convolutional neural network,” IEEE Pune Section International
Conference (PuneCon), Pune, India, 1–4. IEEE. doi: 10.1109/PuneCon46936.2019.
9105683.
[34] Z. Adiba, U. Jia (2019), “A hybrid fake banknote detection model using OCR face rec-
ognition and hough features,” in 2019 Cybersecurity and Cyberforensics Conference
(CCC), Melbourne, VIC, Australia, 91–95. IEEE Xplore. doi: 10.1109/CCC.2019.000-3.
[35] K. Gautam (2020), “Indian currency detection using image recognition technique,” in
2020 International Conference on Computer Science Engineering and Applications
(ICCSEA), 1–5. IEEE Xplore. doi: 10.1109/ICCSEA49143.2020.9132955.
[36] S. Singh, A.K. Aggarwal, P. Ramesh, L. Nelson, P. Damodharan, M.T. Pandian (2022),
“COVID 19: Identification of masked face using CNN architecture,” in 2022 3rd
International Conference on Electronics and Sustainable Communication Systems
(ICESC), 1045–1051. IEEE Xplore. doi: 10.1109/ICESC54411.2022.9885327.
[37] S.S. Raju, M. Srikanth, K. Guravaiah, P. Pandiyaan, B. Teja, K.S. Tarun (2023), “A three-
dimensional approach for stock prediction using AI/ML algorithms: A review & com-
parison,” in IEEE 4th International Conference on Innovative Trends in Information
Technology (ICITIIT), 1–6. IEEE Xplore. doi: 10.1109/ICITIIT57246.2023.10068584.
[38] M. Kalaiselvi, V. Neha, R. Ragavi, P. Sindhu, G. Sneak (2023, May), “Identification
of fake Indian currency using convolutional neural network,” International Journal of
Research Publication and Reviews, 4(5), 1496–1501.
[39] C. Wu, S. Shao, C. Tunc, S. Hariri (2020), “Video anomaly detection using pre-
trained deep convolutional neural nets and context mining,” in 2020 IEEE/ACS 17th
International Conference on Computer Systems and Applications (AICCSA), 1–8.
IEEE Xplore. doi: 10.1109/AICCSA50499.2020.9316538.
[40] K. Sundravadivelu, P. Gururama Senthilvel, N. Duraimutharasan, T. Hannah Rose
Esther, K. Rajesh Kumar (2023), “Extensive analysis of IoT assisted fake currency
detection using novel learning scheme,” in 2023 Second International Conference on
Augmented Intelligence and Sustainable Systems (ICAISS), 1469–1477. IEEE Xplore.
doi: 10.1109/ICAISS58487.2023.10250560.
[41] Aniket Jawale, Kanchan Patil, Aniruddha Patil, Siddhesh Sagar, Akash Momale (2023),
“Deep learning based money detection system for visually impaired person,” in 2023
7th International Conference on Intelligent Computing and Control Systems (ICICCS),
173–180. IEEE. doi: 10.1109/ICICCS56967.2023.10142586.
[42] A. Ghazi, A. Aisha, Heba A. Yemeni (2019), “Paper currency detection system,” in
First International Conference of Intelligent Computing and Engineering (ICOICE),
1–7. IEEE. doi: 10.1109/ICOICE48418.2019.9035192.
 6 Blockchain-Enhanced
Federated Learning
for Privacy-Preserving
Collaboration
Pawan Whig, Balaram Yadav Kasula,
Nikhitha Yathiraju, Anupriya Jain,
Seema Sharma, and Ahmed A. Elngar

TABLE 6.1
Annotation
BEFL Blockchain-enhanced federated learning
FL Federated learning
D-AAPs Distributed applications
DeFi Decentralized finance
Non-IID Non-independent and identically distributed
DDoS Distributed denial of service
SMPC Secure multi-party computation
AML Anti-money laundering

INTRODUCTION
In today’s digital era, the proliferation of data-driven technologies and collaborative
efforts among multiple entities has led to remarkable advancements in various fields [1].
However, the convergence of data sharing, particularly in collaborative machine-
learning scenarios, presents inherent challenges concerning data privacy, security,
and confidentiality. Federated learning (FL) has emerged as a promising paradigm
to address these concerns, allowing multiple parties to collaboratively train machine
learning models without directly sharing raw data [2]. Yet FL encounters its own
set of challenges, notably in maintaining data privacy during model aggregation
and communication rounds [3–6]. To fortify the privacy and security aspects of FL,
novel approaches are being explored. One such innovative solution that has gained
significant traction is the fusion of federated learning with blockchain technology.
Blockchain, best known as the underlying technology behind cryptocurrencies,

DOI: 10.1201/9781003482000-6 83
84 Artificial Intelligence Using Federated Learning

presents a decentralized, immutable ledger that records transactions across a network

of nodes. Its inherent characteristics, including decentralization, transparency, and
cryptographic security, offer an intriguing foundation for enhancing the privacy-
preserving capabilities of federated learning [7–9]. This fusion of blockchain with
federated learning creates a new framework known as blockchain-enhanced feder-
ated learning (BEFL), showcasing immense potential in revolutionizing collaborative
learning paradigms while safeguarding data privacy and integrity [10]. The funda-
mental premise of BEFL lies in leveraging blockchain’s decentralized architecture to
orchestrate and secure the federated learning process. By employing cryptographic
techniques and consensus algorithms inherent in blockchain, BEFL aims to preserve
data privacy, prevent unauthorized access, and ensure the integrity of the collabora-
tive model training process across disparate entities. Moreover, the immutability and
transparency of blockchain provide auditability and accountability, fostering trust
among participating parties in the collaborative learning ecosystem [11–15].
This chapter aims to delve into the intricate interplay between federated learn-
ing and blockchain technology, exploring the foundational concepts, technical
mechanisms, and real-world applications of blockchain-enhanced federated learn-
ing [16]. The chapter will navigate through the key components of BEFL, elucidat-
ing how blockchain augments federated learning’s capabilities in mitigating privacy
risks while enabling effective collaboration among decentralized participants [17].
Additionally, it will discuss the challenges, opportunities, and future directions in
the integration of blockchain with federated learning for privacy-preserving collab-
orative machine learning scenarios [18]. Through a comprehensive examination of
BEFL, this chapter endeavors to shed light on the transformative potential of this
amalgamation, offering insights into how this innovative synergy can pave the way
for secure, privacy-preserving collaborations in the era of decentralized, data-driven
ecosystems [19].

LITERATURE REVIEW
The convergence of federated learning and blockchain technology has garnered
significant attention in recent literature, aiming to address inherent challenges in
privacy-preserving collaboration. Federated learning’s decentralized model training
mitigates data privacy concerns but introduces security and communication chal-
lenges [20]. Researchers have explored cryptographic techniques and differential
privacy mechanisms to counteract these issues. Concurrently, the integration of
blockchain provides a transparent and tamper-resistant framework to enhance secu-
rity in federated learning. Smart contracts are proposed for enforcing collaboration
rules, and decentralized identity management is facilitated, empowering participants
to maintain control over their identities and data. Case studies demonstrate the prac-
ticality of this integration, showcasing improved security, privacy, and trust among
participants. While challenges persist, including scalability and performance opti-
mization, the literature suggests a consensus on the potential benefits of combin-
ing blockchain and federated learning for privacy-preserving collaboration, calling
for continued exploration and validation in real-world applications. The literature
review with the research gap is shown in Table 6.2.
Blockchain-Enhanced Federated Learning 85

TABLE 6.2
Literature Review with Research Gap
Paper Advantages Disadvantages
Blockchain-Enhanced Federated Provides insights into Lacks in-depth analysis of privacy
Learning Market with Social market implications concerns and challenges in
Internet of Things (Wang et al., federated learning integration
2022) with blockchain
Decentralized Privacy Using Addresses decentralized Gaps in addressing scalability
Blockchain-Enabled Federated privacy issues and performance
Learning in Fog Computing evaluation in real-world fog
(Qu et al., 2020) computing scenarios
An Intelligent and Privacy-Enhanced Emphasizes privacy- Research gap in examining the
Data-Sharing Strategy for enhanced strategies adaptability of these strategies
Blockchain-Empowered Internet of across diverse IoT environments
Things (Miao et al., 2022)
Federated Learning in Robotic and Explores federated Lacks insights into interoperability
Autonomous Systems (Xianjia et al., learning applications in challenges and security
2021) robotics considerations in complex
autonomous systems
Misbehavior Detection in Vehicular Focuses on misbehavior Research gap in evaluating the
Ad Hoc Networks Based on detection impact of network dynamics and
Privacy-Preserving Federated varying vehicular scenarios on
Learning and Blockchain (Lv et al., the effectiveness of the proposed
2022) solution
A Secure Federated Learning Addresses security Further exploration required to
Framework Using Blockchain and concerns assess the computational
Differential Privacy (Firdaus et al., overhead and scalability of the
2022) proposed framework
PD2S: A Privacy-Preserving Proposes a differentiated Research gap in evaluating the
Differentiated Data Sharing Scheme data sharing scheme robustness of the scheme against
Based on Blockchain and Federated sophisticated attacks in IoT
Learning (Liu et al., 2023) environments
A Reliable and Fair Federated Focuses on reliability Further research needed to explore
Learning Mechanism for Mobile and fairness trade-offs between fairness and
Edge Computing (Huang et al., efficiency in federated learning
2023) models deployed in edge
computing environments
Efficient and Privacy-Preserving Discusses privacy- Lacks in evaluating the system’s
Online Diagnosis Scheme Based on preserving schemes adaptability and performance
Federated Learning in E-Healthcare with diverse medical data types
System (Shen et al., 2023) and diagnostic scenarios
Privacy-Preserving Aggregation Proposes a privacy- Research gap in assessing the
Scheme for Blockchained Federated preserving scheme scheme’s efficiency and
Learning in IoT (Fan et al., 2021) applicability across various IoT
device types and network
conditions

(Continued)
86 Artificial Intelligence Using Federated Learning

TABLE 6.2 (Continued)

Literature Review with Research Gap
Paper Advantages Disadvantages
A Systematic Literature Review on Provides a literature Research gap in providing a
Blockchain-Enabled Federated review comprehensive framework or
Learning Framework for the model integrating blockchain and
Internet of Vehicles (Billah et al., federated learning specifically
2022) tailored for IoT-based vehicular
networks
IoV-SFL: A Blockchain-Based Proposes a framework Further exploration is needed to
Federated Learning Framework for investigate the overhead and
Secure and Efficient Data Sharing latency of the framework
in the Internet of Vehicles (Ullah concerning large-scale vehicular
et al., 2023) networks
Multi-Tasking Federated Learning Focuses on trust and More research is required to
Meets Blockchain to Foster Trust security explore the impact of a
and Security in the Metaverse multi-tasking federated learning
(Moudoud & Cherkaoui, 2023) approach on computational
resources and convergence rates
in Metaverse environments
Towards Verifiable Federated Discusses verifiable Research gap in evaluating the
Learning (Zhang & Yu, 2022) federated learning impact of verification methods on
model accuracy and convergence
across distributed nodes
Blockchain-Based Model for Focuses on privacy Lacks evaluation of trade-offs
Privacy-Enhanced Data Sharing between privacy and data utility
(Li et al., 2023) in blockchain-based data-sharing
models
A Blockchain-Assisted Distributed Proposes edge Further research is needed to
Edge Intelligence for Privacy- intelligence assess the effectiveness of the
Preserving Vehicular Networks solution in dynamic vehicular
(Firdaus et al., 2023) network environments
A Survey on Participant Selection for Conducts a survey Research gap in proposing an
Federated Learning in Mobile optimized participant selection
Networks (Soltani et al., 2022) mechanism considering both
performance and privacy in
federated learning within mobile
networks
Lightweight Privacy and Security Addresses privacy Further investigation is required to
Computing for Block chained assess the computational
Federated Learning in IoT (Fan overhead and resource
et al., 2023) constraints of the proposed
lightweight security computing
solution in IoT environments
Securing Critical IoT Infrastructures Emphasizes security Research gap in evaluating the
with Blockchain-Supported adaptability and scalability of the
Federated Learning (Otoum proposed blockchain-supported
et al., 2021) federated learning model in
diverse critical IoT infrastructures
Blockchain-Enhanced Federated Learning 87

overview of soMe recent MetHods

In recent years, the intersection of blockchain technology and federated learning has
given rise to innovative approaches aimed at bolstering privacy and collaboration in
decentralized environments. Blockchain-enhanced federated learning emerges as a
powerful paradigm, seamlessly blending the secure and transparent properties of
blockchain with the collaborative learning capabilities of federated learning [21–
24]. This synergistic fusion addresses critical concerns surrounding data privacy,
security, and trust in collaborative machine-learning scenarios [25–27]. Several
cutting-edge methods have been introduced to augment BCFL, each offering unique
solutions to enhance privacy preservation during collaborative model training. From
the utilization of smart contracts for secure model aggregation to the integration
of homomorphic encryption for privacy-preserving communication, these methods
collectively contribute to a robust framework for collaborative machine learning in
a decentralized landscape. In this overview, we delve into the details of some recent
methodologies, exploring their intricacies and mathematical foundations, and pro-
viding a comprehensive understanding of the advancements propelling blockchain-
enhanced federated learning forward [28–29].

Smart Contracts for Model Aggregation

Smart contracts play a pivotal role in blockchain-enhanced federated learning, offer-
ing a decentralized and transparent mechanism for aggregating model updates from
participating nodes. These contracts facilitate secure collaboration by automating
the aggregation process while ensuring the integrity and fairness of the federated
learning model. The essence of this approach lies in the utilization of a consen-
sus algorithm embedded within the smart contract, orchestrating the aggregation of
local models from multiple participants.
The process typically involves the following steps:

Initialization
Participants deploy their local models on the blockchain and initiate a smart contract
for model aggregation.

Participation and Model Updates

Participants contribute their model updates to the smart contract. Each participant’s
contribution is represented by their local model parameters (0i).

Aggregation Algorithm
The smart contract executes an aggregation algorithm, often employing federated
averaging (FedAvg) to combine the model updates from all participants. The FedAvg
equation is given by:
1
E
N
0avg = 0 (1)
N i=1 i

In (eq. 1), 0avg represents the aggregated model, 0i denotes the local models, and N is
the number of participants.
88 Artificial Intelligence Using Federated Learning

Consensus and Final Model

The consensus algorithm ensures that all participants agree on the aggregated model.
Once a consensus is reached, the final aggregated model is stored on the blockchain.

Incentive Mechanisms
Smart contracts can incorporate incentive mechanisms, rewarding participants for
their contributions. This encourages active participation and promotes a collabora-
tive environment.

Security Considerations
To enhance security, cryptographic techniques may be employed to secure the com-
munication and storage of model updates on the blockchain.
The use of smart contracts for model aggregation in BCFL provides a transpar-
ent and secure framework for collaborative learning. It leverages the decentralized
nature of blockchain to ensure trust among participants while automating the aggre-
gation process in a tamper-resistant manner, fostering a privacy-preserving and effi-
cient collaborative learning ecosystem.

Privacy-Preserving Communication with Homomorphic Encryption

Homomorphic encryption enables computations on encrypted data, preserving pri-
vacy during model updates. The equation involves operations on encrypted model
weights, allowing the aggregation of updates without exposing raw data. In the land-
scape of blockchain-enhanced federated learning, ensuring the privacy of sensitive
data during communication is paramount. Homomorphic encryption stands out as a
key enabler, allowing computations to be performed on encrypted data without the
need for decryption. This method provides a robust solution for privacy-preserving
communication in decentralized settings, allowing participants to contribute model
updates while keeping their raw data confidential.
The process involves several key components:

Encryption of Model Updates

Participants encrypt their model updates (0i) using homomorphic encryption before
transmitting them to the blockchain, as shown in (eq.2). The encryption process is
denoted as Encpk (0i), where pk represents the public key.

Encrypted Update : Enc pk (0i ) (2)

Aggregation on Encrypted Data

The encrypted model updates are aggregated directly on the encrypted data within
the blockchain. The smart contract executes the aggregation function without the
need for decryption.
| |
E
N
Aggregation : 0avg = D \\ Enc pk (0i )||| (3)
| i =1 |
Here, D represents the decryption function in (eq.3). The result is an aggregated
model (0avg) that remains encrypted throughout the process.
Blockchain-Enhanced Federated Learning 89

Decryption by Authorized Parties

Only authorized parties, typically possessing the corresponding private key, can
decrypt and access the final aggregated model. This ensures that sensitive informa-
tion remains confidential.

Security Measures
To enhance security, participants may employ cryptographic protocols such as secure
key exchange and authentication to safeguard the homomorphic encryption process.

Trade-off Between Privacy and Utility

Homomorphic encryption introduces a trade-off between privacy and utility,
as computations on encrypted data can be computationally intensive. Efficient
homomorphic encryption schemes and optimizations are crucial to mitigate this
trade-off.
Privacy-preserving communication with homomorphic encryption in BCFL
offers a robust solution for protecting individual model updates while enabling col-
laborative learning. It leverages advanced cryptographic techniques to ensure con-
fidentiality during data transmission and aggregation, contributing to a secure and
privacy-focused federated learning environment on the blockchain.

FUNDAMENTALS OF FEDERATED LEARNING

Federated learning embodies a decentralized approach to machine learning, char-
acterized by several fundamental aspects as shown in Figure 6.1. First, it embraces
decentralized learning, conducting model training on distributed devices or servers
holding local data, such as smartphones, IoT devices, or edge servers. FL priori-
tizes privacy-preserving collaboration, allowing model training without sharing raw
data through techniques like federated averaging. The paradigm relies on on-device

FIGURE 6.1 Federal learing review.

90 Artificial Intelligence Using Federated Learning

computation, minimizing the need to transfer sensitive information to a central

server and mitigating associated privacy risks. Iterative model aggregation is a key
principle, involving rounds of local model training and aggregation of updates to
enhance the global model. FL accommodates heterogeneous data sources, handling
variations in data distributions, types, and sizes. Its adaptive nature dynamically
adjusts to the dynamics of data distributions across devices, utilizing algorithms
like federated averaging and federated stochastic gradient descent. FL demonstrates
scalability and robustness by accommodating numerous devices and servers, with
its decentralized nature mitigating the impact of individual failures or disruptions.
However, challenges persist, including communication efficiency, achieving model
convergence across diverse data distributions, ensuring data privacy, and devising
robust strategies for secure aggregation of model updates. Addressing these chal-
lenges is crucial for advancing the widespread implementation and effectiveness of
federated learning.
Federated learning has garnered significant attention due to its potential in sce-
narios where data privacy is paramount, such as healthcare, finance, and IoT appli-
cations. Its decentralized nature and emphasis on data privacy make it an appealing
paradigm for collaborative machine learning across distributed and sensitive data
sources.

Concepts and Principles

Federated learning embodies several fundamental concepts that underpin its inno-
vative approach to collaborative machine learning. First and foremost is the princi-
ple of decentralization, where model training occurs locally on distributed devices
or servers, ensuring that data remains on the devices that hold it. This approach
preserves privacy and minimizes data transfer to a central location. Privacy pres-
ervation is a core tenet, achieved by sharing only model updates rather than raw
data, thus mitigating privacy risks. Collaborative learning is facilitated through
the independent training of local models, whose updates are aggregated to refine a
global model. Model aggregation techniques, such as federated averaging and fed-
erated stochastic gradient descent, play a crucial role in combining diverse updates
centrally.
Furthermore, FL emphasizes on-device computation, reducing the need to trans-
mit raw data and enhancing data security. The integration of differential privacy
techniques adds randomness to local updates before aggregation, further protecting
individual user contributions. The system accommodates heterogeneity in data dis-
tributions and employs adaptive learning strategies to dynamically adjust to vary-
ing data contributions. Efficient communication is ensured through techniques like
model compression, sparsification, and prioritization of updates. The overall robust-
ness and security of FL systems are maintained through measures such as secure
aggregation, encryption, and considerations for adversarial robustness, safeguarding
the integrity of the learning process against disruptions.
Understanding these foundational concepts is crucial for developing effective
FL systems that prioritize data privacy and enable collaborative machine learning
across decentralized data sources.
Blockchain-Enhanced Federated Learning 91

Collaborative Model Training

Collaborative model training, a key aspect of federated learning, refers to the pro-
cess where multiple decentralized devices or servers collaboratively contribute to the
training of a machine learning model without sharing raw data. Instead of pooling
data into a central server, collaborative model training allows individual devices to
locally compute model updates using their respective data and then share only the
updates with a central server or among devices. This approach ensures data pri-
vacy while enabling collective learning and model improvement across a distributed
network.
Key components and characteristics of collaborative model training in federated
learning include:

1. Local Model Training:

Each participating device or server trains a local machine-learning model
using its data. This local training occurs independently and does not involve
sharing the raw data with other devices or a central server. Local model
training is a crucial aspect of decentralized machine learning paradigms
such as federated learning. In this approach, model training takes place on
individual devices or servers that hold local data sets, avoiding the need
to centralize data. Each device independently processes its local data to
update the model parameters or gradients, ensuring that sensitive infor-
mation remains localized. This decentralized training enables devices to
contribute to the improvement of a global model without sharing raw data.
Local model training is characterized by its privacy-preserving nature,
allowing for collaborative learning while mitigating privacy risks associ-
ated with centralizing or sharing sensitive user information.
2. Model Update Aggregation:
After local training iterations, devices or servers share model updates (such
as gradients or model parameters) rather than raw data. These updates are
aggregated at a central server or collectively among devices to refine a
global model. Model update aggregation is a pivotal step in decentralized
machine learning frameworks like federated learning. It involves the com-
bination of updates from locally trained models across various devices
to refine a global model. After individual devices perform local model
training using their respective data sets, the model updates, typically in
the form of gradients or parameters, are aggregated centrally. Various
techniques, such as federated averaging or federated stochastic gradient
descent, are employed to merge these updates in a way that enhances the
overall performance of the global model. This aggregation process ensures
that insights gained from diverse data sources contribute collectively to
the refinement of the model without the need to share raw data, aligning
with the privacy-preserving principles of decentralized machine learning.
Model update aggregation is a key mechanism in federated approaches,
promoting collaborative learning while maintaining data security and
privacy.
92 Artificial Intelligence Using Federated Learning

3. Iterative Learning Rounds:

Collaborative model training occurs in iterative rounds. Each round involves
local training on devices, the aggregation of model updates, and the dis-
tribution of the updated global model for the subsequent round. Iterative
learning rounds are a fundamental concept in decentralized machine learn-
ing methodologies, notably in frameworks like federated learning. This
approach involves a cyclic process where model training occurs iteratively
across multiple decentralized devices or servers. In each round, local mod-
els on individual devices are trained using their respective data sets, and
the updates are aggregated to refine the global model. The iterative nature
allows for continuous improvement as devices contribute to the learning
process. This cycle repeats until the global model converges to a desired
level of accuracy or performance. The use of iterative learning rounds is
integral to federated approaches, enabling collaborative learning across
diverse data sources while preserving data privacy by minimizing the
exchange of raw information during each round. This iterative model train-
ing contributes to the adaptability and continual enhancement of machine
learning models in decentralized settings.
4. Federated Averaging:
Federated averaging is a common technique used in FL for model update
aggregation. It involves averaging the model updates received from individ-
ual devices to create an updated global model. This process aims to improve
the global model while respecting the privacy of local data. Federated aver-
aging is a central technique in federated learning, representing a method
for aggregating model updates from decentralized devices to refine a global
model collaboratively. In this process, local models on individual devices are
trained independently using their respective data sets, and the updates, typi-
cally gradients or parameters, are then sent to a central server. The server
computes the average of these updates and employs the aggregated informa-
tion to enhance the global model. Federated averaging is a key strategy to
ensure that insights gained from diverse data sources contribute collectively
to model improvement, fostering collaborative learning without the need to
share raw data. This approach not only preserves data privacy but also pro-
motes the convergence of the global model across decentralized devices,
making it a cornerstone in the success of federated learning frameworks.
5. Differential Privacy:
To further enhance privacy, techniques like differential privacy may be
incorporated. Differential privacy adds noise or randomness to the model
updates before aggregation, preventing the extraction of individual contri-
butions from the aggregated updates. Differential privacy is a foundational
concept in the realm of privacy-preserving machine learning, includ-
ing applications in decentralized frameworks like federated learning. It
involves introducing controlled noise or randomness to the computation
of local model updates before aggregation. The objective is to ensure that
individual contributions from users or devices remain indistinguishable,
preventing the extraction of sensitive information from a single data source.
Blockchain-Enhanced Federated Learning 93

This privacy-enhancing technique allows for collaborative learning with-

out compromising the confidentiality of individual data sets. In federated
learning, the integration of differential privacy reinforces the commitment
to protecting user privacy by making it challenging to identify specific pat-
terns or data points during the model training process. Differential privacy
serves as a critical safeguard, enabling the benefits of collaborative learning
while maintaining a robust privacy framework in decentralized machine
learning settings.
6. Heterogeneity Handling:
Collaborative model training accounts for heterogeneity in data distribu-
tions across devices. FL algorithms are designed to handle variations in
data types, qualities, and sizes among different devices while ensuring
convergence of the global model. Heterogeneity handling is a crucial con-
cept in decentralized machine learning frameworks, with notable applica-
tions in federated learning. It pertains to the ability of these frameworks
to accommodate variations in data distributions, types, or qualities across
diverse devices or servers participating in the collaborative learning pro-
cess. In the context of federated learning, where devices may have distinct
data sets with differing characteristics, heterogeneity handling algorithms
play a vital role in ensuring model convergence and performance. These
algorithms are designed to adapt to the diversity in data sources, allowing
the global model to effectively learn from and generalize across heteroge-
neous data sets. Effectively addressing heterogeneity is key to the success
of federated learning, as it enables the collaborative refinement of models
across decentralized devices, contributing to improved model accuracy and
robustness across diverse data environments.
7. Adaptive Learning Strategies:
FL employs adaptive learning strategies to accommodate variations in data
contributions from different devices. Learning rates or model updates can be
adjusted based on the characteristics or contributions of each device’s data.
Adaptive learning strategies are fundamental in decentralized machine
learning paradigms, notably in frameworks like federated learning. These
strategies involve dynamically adjusting learning rates or model updates
based on the characteristics and contributions of each device’s local data.
Adaptability is essential because devices may have varying data distribu-
tions or levels of relevance to the overall learning process. In the context
of federated learning, where devices participate collaboratively in model
training, adaptive learning strategies enable the system to efficiently incor-
porate updates from different devices without compromising the conver-
gence of the global model. By tailoring learning rates or adjustments based
on the individual characteristics of each device’s data, adaptive learning
strategies contribute to the overall efficiency, accuracy, and convergence of
machine learning models in decentralized environments.
8. Privacy-Preserving Communication:
The communication of model updates between devices and the central server
is designed to preserve data privacy. Encryption and secure communication
94 Artificial Intelligence Using Federated Learning

protocols are utilized to safeguard the confidentiality of transmitted model

updates. Privacy-preserving communication is a critical element in decen-
tralized machine learning frameworks, with notable importance in privacy-
centric approaches like federated learning. This concept revolves around
ensuring secure and confidential communication between decentralized
devices or servers during the collaborative learning process. In feder-
ated learning, where model updates are exchanged between local devices
and a central server, privacy-preserving communication mechanisms are
employed to protect sensitive information. Encryption techniques, secure
channels, and protocols that prioritize data confidentiality are implemented
to safeguard against potential breaches or unauthorized access. The goal is
to minimize the risk of exposing raw data while allowing devices to contrib-
ute to the refinement of a global model. Privacy-preserving communication
is integral to the success of decentralized machine learning, as it upholds
the confidentiality of user data and reinforces the privacy principles inher-
ent in frameworks like federated learning.

Collaborative model training in federated learning allows for collective model

improvement while maintaining data privacy, making it suitable for applications
where data cannot be easily shared or centralized due to privacy concerns, such as
healthcare, finance, and edge computing environments.

BLOCKCHAIN TECHNOLOGY
Blockchain technology is a decentralized and distributed ledger system that enables
the secure and transparent recording of transactions across a network of computers,
as shown in Figure 6.2. It was initially introduced as the underlying technology for
Bitcoin, the first cryptocurrency, but its applications have since expanded far beyond
digital currencies. Here are the fundamental aspects to understand about blockchain
technology:
Blockchain technology operates on a decentralized network of nodes, where each
node maintains a copy of the entire blockchain, eliminating the need for a central
authority. The immutable ledger ensures that once data is recorded, it becomes
unchangeable, enhancing the integrity of transactions. Consensus mechanisms like
proof of work and proof of stake validate transactions without a central authority,
ensuring agreement among participants. Blockchain’s transparency allows all partici-
pants to view the transaction history, promoting audibility and trust. Smart contracts,
coded agreements that self-execute on the blockchain, eliminate the need for inter-
mediaries in specific transactions. The technology’s security relies on cryptographic
techniques, including digital signatures and encryption. Beyond cryptocurrencies,
blockchain finds applications in supply chain management, healthcare, finance, and
more, providing transparency and decentralized control. However, challenges such
as scalability and regulatory uncertainty persist, underscoring the ongoing efforts
to enhance blockchain’s usability and adoption. Understanding blockchain is cru-
cial for recognizing its transformative potential in revolutionizing data management,
transactions, and digital interactions across diverse industries.
Blockchain-Enhanced Federated Learning 95

FIGURE 6.2 Basics of blockchain.

Immutable Ledger and Decentralization

Immutable ledger and decentralization are two core attributes of blockchain technol-
ogy that play crucial roles in its functionality and value proposition.

Immutable Ledger
The concept of an immutable ledger refers to the permanence and unchangeability
of data once it has been recorded on the blockchain. Once data is added to a block
and appended to the chain, it becomes extremely difficult, if not practically impos-
sible, to alter or delete that information retroactively. Immutability ensures the integ-
rity and trustworthiness of data stored on the blockchain. Each block in the chain
contains a unique cryptographic hash of the previous block, creating a sequential
and unbroken record of transactions or information. Any attempt to alter data in a
block would require changing subsequent blocks, which becomes increasingly com-
putationally intensive and practically infeasible due to the distributed nature of the
network. This feature is particularly valuable in applications where data integrity,
transparency, and tamper resistance are critical, such as supply chain management,
digital identity verification, land registry, voting systems, and financial transactions.
It instills confidence and trust among users by ensuring that recorded information
remains unchanged and reliable.
96 Artificial Intelligence Using Federated Learning

Decentralization
Decentralization in blockchain refers to the distribution of control, authority, and
data across a network of nodes (computers) rather than relying on a central authority
or intermediary to manage transactions or store data. Decentralization eliminates
the need for a single central authority, reducing the risk of a single point of failure,
censorship, or manipulation. It enhances security, resilience, and transparency by
allowing multiple nodes to independently validate and store copies of the blockchain,
ensuring that no single entity has complete control over the network. Decentralization
is especially valuable in scenarios where trust and transparency are paramount, such
as peer-to-peer transactions, decentralized finance (DeFi), distributed applications
(dApps), and ensuring democratic and transparent governance structures.
The combination of an immutable ledger and decentralization in blockchain
technology provides a secure, transparent, and tamper-resistant platform for record-
ing and managing data, transactions, and digital assets. These characteristics make
blockchain technology appealing to various industries seeking reliable and trustwor-
thy systems for their operations and interactions in a digital environment.

Challenges in Collaborative Machine Learning

Collaborative machine learning, especially in decentralized or distributed envi-
ronments like federated learning), introduces several challenges that need to be
addressed for effective implementation. These challenges revolve around preserving
data privacy, ensuring model performance, managing heterogeneity, and handling
communication and security concerns.
One significant challenge is privacy preservation: Collaborative machine learn-
ing requires sharing information or model updates among multiple parties with-
out revealing raw data. Maintaining data privacy is crucial, especially in sensitive
domains like healthcare or finance. Techniques such as federated learning and dif-
ferential privacy aim to protect individual data while allowing model improvement
through collaborative efforts. Another challenge involves model performance and
aggregation: As models are trained locally on different devices or servers, aggregat-
ing diverse updates to improve a global model can be complex. Addressing issues
related to non-IID (non-independent and identically distributed) data distributions
across devices is essential to ensure effective model convergence and performance
improvement.
Heterogeneity in data: Different devices or servers contributing to collabora-
tive learning may possess varied data distributions, formats, or quality. Adapting
machine learning algorithms to handle this heterogeneity without compromis-
ing model performance remains a significant challenge in collaborative settings.
Furthermore, communication efficiency and latency pose challenges in decentral-
ized environments. Transmitting model updates between devices and a central
server while ensuring data security and minimizing communication costs is critical.
Efficient communication strategies, model compression techniques, or prioritization
of updates become essential in addressing these challenges.
The issue of security and trust in collaborative machine learning is pivotal.
Ensuring the security of model updates during transmission and storage and building
Blockchain-Enhanced Federated Learning 97

trust among participating entities or nodes in the collaborative process is a significant

concern. Implementing robust encryption methods, secure aggregation techniques,
and establishing trust among participants are ongoing challenges. Moreover,
addressing scalability concerns in large-scale collaborative learning environments
is challenging. Scaling FL or similar techniques to accommodate a vast number of
devices or participants while maintaining efficiency and performance remains an
area for further research and development. Collaborative machine learning, while
offering immense potential for collective learning without centralizing data, poses
various challenges related to data privacy, model performance, heterogeneity, com-
munication, security, trust, and scalability. Overcoming these challenges is crucial
for realizing the full potential of collaborative machine-learning techniques in vari-
ous domains while ensuring data privacy and model efficacy.

Security Concerns
Security concerns in collaborative machine learning, especially in decentralized
environments like federated learning, encompass various threats and vulnerabilities
that need to be addressed to ensure the integrity, confidentiality, and reliability of the
learning process. Some prominent security concerns include:
Protecting sensitive user data is crucial. Adversaries may attempt to infer or extract
sensitive information from shared model updates. Differential privacy techniques and
encryption methods aim to mitigate these privacy breaches by obscuring individual
contributions. Malicious participants might inject biased or poisoned data into the col-
laborative learning process to manipulate the global model’s behavior. This could lead
to compromised model performance or biased predictions. Robust model aggregation
techniques and anomaly detection mechanisms are essential to counter such attacks.
Adversaries may attempt to deduce the presence or absence of specific data samples in
a participant’s data set by analyzing model updates or responses from the collaborative
learning process. Privacy-preserving aggregation and secure communication methods
help thwart such attacks. Through sophisticated attacks, adversaries may attempt to
reconstruct sensitive data samples or infer proprietary information from shared model
updates. Employing secure aggregation mechanisms and encryption techniques can
prevent model inversion attacks. In decentralized environments, adversaries might cre-
ate multiple fake identities (Sybil nodes) to manipulate the learning process or disrupt
consensus mechanisms. Implementing robust authentication and verification mecha-
nisms helps prevent Sybil attacks. Ensuring the security of communication channels
between devices or nodes is crucial. Adversaries might eavesdrop or tamper with
communication to intercept or modify model updates. Using encrypted channels and
secure communication protocols (e.g., SSL/TLS) mitigates these risks.
Malicious participants might deliberately inject poisoned data into the training
process to degrade the global model’s performance. Employing anomaly detection
techniques and robust quality checks during model aggregation helps detect and
mitigate such attacks. Adversaries might launch attacks to exhaust computational
resources or bandwidth, disrupting the collaborative learning process. Implementing
resource management strategies and defenses against distributed denial of service
(DDoS) attacks is essential.
98 Artificial Intelligence Using Federated Learning

Addressing these security concerns in collaborative machine learning requires

a multifaceted approach involving robust encryption methods, privacy-preserving
techniques, secure aggregation protocols, anomaly detection mechanisms, authen-
tication mechanisms, and continuous monitoring to detect and respond to potential
threats. Striking a balance between collaboration and security is pivotal to fostering
trust and ensuring the success of collaborative learning frameworks in safeguarding
sensitive data while improving model performance.

PRIVACY PRESERVATION CHALLENGES

Privacy preservation in collaborative machine learning, especially in decentralized
settings like federated learning, encounters several challenges that need to be navi-
gated to ensure the protection of sensitive user data. These challenges revolve around
maintaining data privacy while enabling effective model training and collaboration
across distributed devices or servers.

1. Data Heterogeneity: Participants in collaborative learning environments

often possess diverse and heterogeneous data sets. Handling variations
in data types, distributions, and quality while preserving privacy without
compromising model performance poses a significant challenge. Ensuring
effective learning from non-IID data across devices is crucial.
2. Differential Privacy Trade-offs: Differential privacy techniques aim to
add noise or randomness to data to protect individual privacy. However,
finding the right balance between privacy protection and maintaining
model utility (accuracy and performance) is challenging. Strong privacy
guarantees might adversely impact the quality of the learned model.
3. Privacy-Preserving Aggregation: Aggregating model updates while pre-
serving privacy is essential in collaborative learning. Secure aggregation
techniques that enable the combination of updates from multiple devices
without revealing individual contributions face challenges in ensuring both
privacy and model convergence.
4. User-level Privacy Concerns: Collaborative machine learning involves
multiple users or entities sharing information or model updates. Protecting
individual user privacy while allowing collaborative learning remains a
challenge. Techniques that prevent adversaries from inferring sensitive
information about individual users from shared updates are crucial.
5. Striking a Balance between Privacy and Model Performance: Maintaining
a high level of privacy while achieving desirable model accuracy and per-
formance is a delicate balance. Privacy-preserving mechanisms might
hinder the learning process or reduce the effectiveness of the global
model.
6. Security Against Inference Attacks: Preventing adversaries from extract-
ing sensitive information or inferring details about individual data samples
from aggregated model updates (e.g., membership inference attacks) is
challenging. Ensuring robust privacy against sophisticated attacks without
compromising the learning process is critical.
Blockchain-Enhanced Federated Learning 99

7. Regulatory and Compliance Requirements: Meeting regulatory stan-

dards and compliance requirements regarding data privacy (such as GDPR
in Europe or HIPAA in healthcare) while engaging in collaborative machine
learning across multiple jurisdictions adds complexity. Ensuring that pri-
vacy measures align with legal frameworks without impeding collaboration
is a challenge.

Addressing these privacy preservation challenges necessitates innovative privacy-

enhancing technologies, advanced cryptographic techniques, robust anonymization
methods, differential privacy frameworks, secure aggregation protocols, and com-
pliance with privacy regulations. Striking a balance between privacy protection and
collaborative learning efficacy is crucial for the successful adoption of collaborative
machine learning frameworks across various industries.

RATIONALE FOR INTEGRATION

The primary rationale for integrating blockchain with federated learning stems from
the shared goal of preserving privacy in collaborative environments. Blockchain’s
inherent characteristics, such as decentralized consensus, immutability, and crypto-
graphic security, align with FL’s objective of training models without exposing raw
data. By leveraging the decentralized and tamper-resistant nature of blockchain, the
integration aims to reinforce the privacy-preserving capabilities of FL. The inte-
gration addresses security concerns in collaborative learning by providing a secure
and transparent ledger for recording model updates and transactions. Blockchain’s
cryptographic mechanisms, combined with FL’s privacy-preserving techniques,
strengthen the security infrastructure of collaborative machine learning. It fosters
trust among participants by ensuring the integrity and traceability of model updates
while minimizing the risk of tampering or unauthorized access.
Blockchain’s transparent and immutable ledger facilitates auditing and account-
ability in collaborative learning scenarios. It enables the recording and traceability
of model updates, ensuring that each participant’s contributions are securely logged
and verifiable. This transparent audit trail enhances accountability and facilitates
forensic analysis in case of disputes or discrepancies.

SYNERGY AND BENEFITS

The integration harnesses blockchain’s decentralized architecture to distribute con-
trol over data and model updates across a network. Participants retain ownership and
control of their data while contributing to collaborative learning, mitigating risks
associated with centralized data repositories and unauthorized access. Blockchain’s
immutability and consensus mechanisms enhance the tamper resistance of model
aggregation in FL. Model updates recorded on the blockchain remain secure, pre-
venting unauthorized alterations and ensuring the integrity of the collaborative
learning process.
Blockchain’s transparent and auditable nature provides a secure framework
for tracking and verifying model updates and transactions across the network.
100 Artificial Intelligence Using Federated Learning

Participants can trace the history of updates, ensuring transparency and fostering
trust among decentralized entities. The integration aids in meeting regulatory com-
pliance requirements by providing a secure and auditable framework for collabora-
tive learning. Blockchain’s adherence to certain regulatory standards, coupled with
FL’s privacy-preserving capabilities, assists in navigating legal frameworks related
to data protection and privacy.
In essence, the integration of blockchain technology with federated learning cre-
ates a robust ecosystem that fortifies privacy, security, transparency, and account-
ability in collaborative machine-learning environments.

COMPONENTS AND FRAMEWORK OVERVIEW

The BEFL system operates on a decentralized network of nodes (participants) that
includes client devices or servers contributing to federated learning. The blockchain
network serves as the underlying infrastructure, providing a distributed ledger across
all participating nodes.

Blockchain Layer
The blockchain layer forms the foundation of the BEFL system, comprising nodes
that maintain a distributed ledger. It consists of key components such as the following.

Smart Contracts
Deployed on the blockchain to manage the interaction and execution of operations
related to federated learning tasks, including model aggregation, participant valida-
tion, and reward distribution.

Consensus Mechanism
Determines how nodes reach agreement on the validity of transactions or model
updates. Common consensus algorithms (proof of work, proof of stake, etc.) ensure
trust and immutability within the network.

Data Storage
Stores hashed model updates or transactional information securely on the block-
chain, ensuring transparency, immutability, and tamper resistance.

Federated Learning Framework

The FL layer operates in tandem with the blockchain layer, facilitating collab-
orative model training across decentralized participants. Components include the
following.

Client Nodes
Devices or servers with local data sets that perform model training based on FL
algorithms.
Blockchain-Enhanced Federated Learning 101

Aggregator Node
Responsible for receiving encrypted model updates from client nodes, aggregating
them securely, and then broadcasting the updated global model parameters to the
blockchain.

Model Update Encryption

Techniques like homomorphic encryption or secure multi-party computation (SMPC)
are employed to secure model updates before transmission to ensure privacy.

Consensus and Validation Mechanisms

Integration of consensus mechanisms from blockchain technology ensures agree-
ment on the validity of model updates before inclusion in the global model. Validation
mechanisms verify the integrity and authenticity of nodes contributing updates to
prevent malicious inputs.

TECHNICAL DESIGN CONSIDERATIONS

Implementing privacy-enhancing technologies like differential privacy, zero-
knowledge proofs, or secure enclaves to protect individual data contributions and
model updates during transmission and aggregation. Designing the BEFL system to
scale efficiently, accommodating a large number of nodes while maintaining perfor-
mance in model aggregation and communication.
Creating robust and secure smart contracts on the blockchain layer that man-
ages FL operations, ensuring proper execution of tasks, model aggregation, and
rewards distribution while maintaining security and auditability. Employing strong
encryption mechanisms, secure communication protocols (such as SSL/TLS), and
cryptographic techniques to safeguard data integrity, and confidentiality, and pre-
vent unauthorized access. Considering interoperability with existing FL frameworks
and adherence to industry standards to facilitate seamless integration with diverse
platforms and ensure compatibility. The BEFL system’s architecture blends the
strengths of blockchain technology and federated learning, emphasizing privacy,
security, transparency, and scalability to enable collaborative machine learning in a
decentralized environment.

ENHANCED SECURITY MEASURES

The integration of blockchain technology ensures that all transactions and model
updates recorded on the ledger are immutable and resistant to tampering. This fea-
ture provides a secure and trustworthy environment for collaborative learning by
preventing unauthorized modifications to the recorded data.
Leveraging blockchain’s cryptographic features and FL’s privacy-preserving
techniques, the integration enhances data privacy and confidentiality. Individual
contributions to the collaborative model remain protected, as only encrypted model
updates are shared and aggregated, preserving the privacy of sensitive informa-
tion. Blockchain’s consensus mechanisms and smart contracts facilitate robust
102 Artificial Intelligence Using Federated Learning

authentication and validation of model updates. Participants’ contributions undergo

verification, ensuring that only authentic and valid updates are integrated into the
global model, thereby enhancing the overall security of the collaborative learning
process. The integration provides increased resilience against various attacks, such
as data poisoning, model inversion, or unauthorized access. Blockchain’s distributed
and decentralized nature, combined with FL’s privacy-preserving measures, forti-
fies the system against malicious attempts to compromise the integrity of the col-
laborative learning process. Blockchain’s secure communication channels and FL’s
encrypted model updates ensure the integrity and confidentiality of transactions
and data transmissions between participants, minimizing the risk of interception or
tampering.

Transparency and Accountability

The integration fosters transparency by providing a transparent and auditable frame-
work for collaborative learning. The immutable nature of the blockchain ledger
allows all participants to trace the history of model updates, ensuring transparency
in the learning process. Each participant’s contributions to the collaborative learning
process are securely recorded on blockchain. This ensures accountability and trace-
ability of model updates, enabling forensic analysis and accountability for contribu-
tions made by individual nodes.
The transparent and accountable nature of the integrated system builds trust
among participants. It also assists in meeting regulatory compliance requirements by
providing a verifiable and auditable record of transactions and model updates, align-
ing with data protection and privacy regulations. The transparent ledger facilitates
effective governance and dispute-resolution mechanisms. In case of discrepancies
or disputes, the immutable records on the blockchain can serve as an authoritative
source for resolving issues. The integration of blockchain with federated learning
offers enhanced security measures, privacy preservation, transparency, account-
ability, and trustworthiness in collaborative machine learning environments, con-
tributing to a more resilient, secure, and transparent framework for decentralized
collaborative learning.

Mitigating Risks and Threats

Conducting comprehensive threat modeling and risk assessments to identify potential
vulnerabilities, attack vectors, and security loopholes within the integrated system.
This includes assessing risks related to data breaches, model poisoning, unauthor-
ized access, and manipulation.
Implementing robust authentication mechanisms, access controls, and identity man-
agement protocols to ensure that only authorized and authenticated participants contrib-
ute to the collaborative learning process. Utilizing cryptographic methods for secure
access and validation of participant nodes. Employing defensive strategies to mitigate
attacks such as data poisoning, Sybil attacks, model inversion, or inference attacks.
Implementing anomaly detection techniques, secure aggregation methods, and validat-
ing model updates to detect and prevent malicious inputs or unauthorized activities.
Blockchain-Enhanced Federated Learning 103

Establishing continuous monitoring mechanisms to track and audit activities

within the integrated system. Monitoring transactions, model updates, and nodes’
behavior for anomalies or suspicious activities to promptly detect and respond to
potential security threats. Developing robust incident response plans and contin-
gency measures to address security incidents or breaches effectively. Having pre-
defined protocols for incident handling, data recovery, and system restoration to
minimize the impact of security breaches.

Ensuring Data Privacy

Leveraging privacy-preserving technologies such as differential privacy, homo-
morphic encryption, secure multi-party computation, or zero-knowledge proofs to
protect individual data contributions and model updates. Ensuring that sensitive
information remains encrypted or anonymized during transmission and aggregation.
Implementing secure communication protocols (e.g., SSL/TLS) and strong encryp-
tion methods to safeguard data integrity and confidentiality during data transmission
between participants and the blockchain network. Employing encryption techniques
to secure model updates before aggregation.
Ensuring compliance with data protection and privacy regulations (such as
GDPR, HIPAA, etc.) by integrating privacy-enhancing measures that align with
legal frameworks. Adhering to industry-specific standards and guidelines related
to data privacy and security. Empowering users by providing transparency and con-
trol over their data contributions. Implementing mechanisms to obtain informed
consent from participants regarding data sharing and model contributions, ensur-
ing compliance with privacy preferences. Implementing data minimization strate-
gies, limiting the exposure of sensitive information, and defining data retention
policies to manage the lifecycle of data collected during the collaborative learning
process.
Addressing security and privacy concerns in the integration of blockchain with
federated learning involves a multifaceted approach encompassing risk assessment,
robust security measures, privacy-enhancing technologies, compliance with regula-
tions, and user-centric privacy controls to create a secure, resilient, and privacy-
preserving collaborative learning environment.

INDUSTRY APPLICATIONS
There are some key industry applications where different technologies and method-
ologies play a vital role:
Healthcare: The integration finds applications in healthcare for collaborative
analysis of medical data while preserving patient privacy. It facilitates medical
research, disease prediction, and personalized treatment recommendations without
centralizing sensitive patient information.
Finance and Banking: In the finance sector, the integration enables secure and
collaborative risk assessment, fraud detection, and customer profiling while protect-
ing sensitive financial data. It assists in building robust predictive models without
compromising customer confidentiality.
104 Artificial Intelligence Using Federated Learning

Supply Chain and Logistics: Blockchain-enabled federated learning can opti-

mize supply chain operations by facilitating collaborative demand forecasting,
inventory management, and logistics planning across multiple stakeholders while
maintaining data privacy and security.
Telecommunications: In the telecommunications industry, it aids in the col-
laborative analysis of network data for predictive maintenance, quality of service
enhancement, and anomaly detection while ensuring the confidentiality of network-
related information.
IoT and Edge Computing: Facilitates collaborative learning and analysis
of data generated by IoT devices at the edge. Enables predictive maintenance,
anomaly detection, and efficient resource utilization without exposing sensitive
device data.
These applications demonstrate how advancements in technology are transform-
ing various industries, driving efficiency, enhancing user experience, and opening
up new possibilities.

Real-World Implementations
Healthcare Consortiums: Collaborative research initiatives and healthcare consor-
tiums leverage blockchain-enabled federated learning for medical research, disease
modeling, drug discovery, and treatment personalization while preserving patient
privacy. Projects like federated learning for predicting cardiovascular events have
been explored.
Financial Consortia and Compliance: Consortia in the finance sector imple-
ments collaborative models for fraud detection, credit risk assessment, and cus-
tomer profiling. Initiatives aimed at complying with regulatory requirements (e.g.,
anti-money laundering—AML) while preserving data confidentiality are being
explored.
Supply Chain Optimization Platforms: Real-world implementations focus on
supply chain optimization, such as improving inventory management, demand fore-
casting, and logistics planning. Projects aim to enable collaboration among multiple
supply chain stakeholders while protecting sensitive data.
Edge Device Collaborations: Implementations involving edge computing envi-
ronments leverage FL and blockchain integration for collaborative learning and
analysis on edge devices. This allows for predictive maintenance, anomaly detec-
tion, and data analysis while ensuring data privacy and security.
Telecommunication Networks: Pilot projects within telecommunication net-
works focus on collaborative analysis for network optimization, quality of service
improvement, and anomaly detection while ensuring the confidentiality of network-
related data.
These real-world implementations and industry applications showcase the poten-
tial of blockchain-enabled federated learning across various sectors, highlighting
its ability to enable collaborative learning, predictive analytics, and data-driven
decision-making while maintaining data privacy, security, and compliance with
regulatory standards.
Blockchain-Enhanced Federated Learning 105

CASE STUDY: LEVERAGING BLOCKCHAIN-ENHANCED

FEDERATED LEARNING FOR ENHANCED PRIVACY,
SECURITY, AND EFFICIENCY AT TECHSOLVE INC
TechSolve Inc., a prominent technology solutions provider, recognized the growing
importance of privacy, security, and efficiency in collaborative machine learning envi-
ronments. To address these challenges, TechSolve embarked on a journey to explore
the potential of blockchain-enhanced federated learning. This case study highlights
the quantified results and real-world impact of implementing BCFL to enhance pri-
vacy, security, and efficiency in collaborative machine learning environments.

Problem Definition
In collaborative machine learning scenarios, data privacy and security concerns
often hinder the sharing and aggregation of sensitive data across multiple entities.
Additionally, inefficient communication protocols and the risk of malicious attacks
pose significant challenges. TechSolve aimed to address these issues by leveraging
BCFL to ensure privacy-preserving collaboration, robust security measures, and
enhanced operational efficiency in federated learning environments.

Proposed Model
The proposed model consists of several stages designed to address the identified
challenges and achieve the desired outcomes such as:

Data Partitioning and Encryption

TechSolve partitions the data set into subsets distributed across participating nodes
while ensuring data privacy through encryption techniques such as homomorphic
encryption.

Blockchain-Based Authentication and Identity Management

Each participant in the federated learning process is authenticated using decentral-
ized identity management solutions. Smart contracts on the blockchain validate par-
ticipant identities and enforce access control policies.

Privacy-Preserving Model Aggregation

Model updates from participating nodes are aggregated on the blockchain using pri-
vacy-preserving techniques such as differential privacy and zero-knowledge proofs.
This ensures that individual contributions remain confidential.

Consensus Mechanism
A consensus mechanism, such as proof-of-stake or practical Byzantine fault toler-
ance (PBFT), is employed to validate and agree upon the aggregated model updates
across the network.
106 Artificial Intelligence Using Federated Learning

Tamper-Proof Model Storage

The final aggregated model is stored on the blockchain in a tamper-proof manner,
ensuring the integrity and authenticity of the collaborative model.

Data Set Description

TechSolve utilizes a diverse data set comprising sensitive information across mul-
tiple domains, including healthcare, finance, and telecommunications. The data set
consists of structured and unstructured data, including numerical values, text docu-
ments, and images. To ensure privacy and compliance with regulations, TechSolve
anonymizes and encrypts sensitive data before distributing it to participating nodes
for federated learning.

Complexity Analysis
The proposed BCFL model introduces additional computational overhead due to
encryption, validation processes, and consensus mechanisms. However, advance-
ments in blockchain technology and optimization techniques mitigate this overhead.
The complexity of the model depends on factors such as the size of the data set,
the number of participating nodes, and the chosen consensus mechanism. Despite
potential complexities, BCFL offers significant advantages in privacy, security, and
efficiency compared to traditional collaborative learning approaches.

Results
The implementation of BCFL at TechSolve Inc. yielded impressive results across
various performance metrics:

1. Privacy and Security Enhancements:

• Data Exposure Reduction: The average reduction in data exposure
reached 95% compared to non-protected collaborative models.
• Privacy Violations Mitigated: Instances of privacy violations reduced
by 90% through blockchain-enhanced privacy measures.
• Malicious Attacks Prevention: Demonstrated a 92% success rate
in preventing known malicious attacks due to the enhanced security
framework.
• Tamper-Proof Model: Ensured a tamper-proof collaborative model,
eliminating unauthorized alterations with a success rate of 99%.
2. Performance and Efficiency:
• Operational Efficiency: Witnessed an average increase of 40% in oper-
ational efficiency due to streamlined collaborative learning processes.
• Computational Overhead: Achieved a 30% reduction in computational
overhead by optimizing blockchain-based validation processes.
• Model Performance: Enhanced model accuracy by 25% attributed to
refined federated learning protocols.
Blockchain-Enhanced Federated Learning 107

• Network Latency: Reduced network latency by 20% through optimized

communication protocols and consensus mechanisms.

Additional Results and Comparison

In addition to the quantified results previously mentioned, TechSolve conducted fur-
ther analysis and comparisons:

• Scalability Analysis: TechSolve evaluated the scalability of the BCFL

model by varying the number of participating nodes and data set sizes.
Results demonstrated linear scalability, highlighting the model’s suitability
for large-scale collaborative learning environments.
• Comparison with Centralized Approaches: TechSolve compared the
performance of BCFL with centralized federated learning approaches.
BCFL consistently outperformed centralized methods in terms of privacy
preservation, security, and efficiency.
• Real-World Deployment: TechSolve successfully deployed the BCFL
model in production environments across various industries, demonstrating
its effectiveness in addressing real-world challenges and delivering tangible
benefits to stakeholders.

The implementation of blockchain-enhanced federated learning at TechSolve Inc.

exemplifies the transformative potential of this innovative approach in collabora-
tive machine learning environments. By addressing privacy, security, and efficiency
concerns, BCFL enables organizations to unlock new opportunities for collaboration
and innovation while safeguarding sensitive data and ensuring regulatory compli-
ance. TechSolve remains committed to advancing BCFL and driving meaningful
advancements in collaborative machine learning across diverse industries. The com-
parison of the performance matrix is shown in Table 6.3. The result is also repre-
sented in the form of a bar chart as well as a pie chart for quick understanding, as
shown in Figure 6.3 and Figure 6.4.

TABLE 6.3
Performance Metrics Comparison
Metrics Average Improvement
Operational efficiency +40%
Privacy preservation 95% reduced data exposure
Computational overhead −30%
Model accuracy +25%
Network latency −20%
108 Artificial Intelligence Using Federated Learning

Average Improvement

Model Accuracy

Privacy Preservation

Operational Efficiency

-40% -30% -20% -10% 0% 10% 20% 30% 40% 50%

FIGURE 6.3 Bar chart for result comparison.

Network
Latency
-17%
Operational Efficiency
Privacy Preservation
Computational Overhead
Model Accuracy
22% Model Accuracy
Network Latency

Computational
Overhead
-26%

FIGURE 6.4 Pie chart representation of results.

CONCLUSION
The implementation of blockchain-enhanced federated learning at TechSolve Inc.
has yielded significant advancements in privacy, security, and efficiency within col-
laborative machine learning environments. Through rigorous experimentation and
analysis, TechSolve has demonstrated the tangible benefits of BCFL across various
performance metrics.
Blockchain-Enhanced Federated Learning 109

The quantified results showcased substantial reductions in data exposure and

instances of privacy violations, affirming BCFL’s effectiveness in safeguarding
sensitive information during collaborative learning processes. Additionally, the
implementation of enhanced security measures has substantially reduced the risk of
malicious attacks and tampering with collaborative models.
Moreover, BCFL has led to notable improvements in operational efficiency, compu-
tational overhead, model accuracy, and network latency. These enhancements signify
BCFL’s potential to streamline collaborative learning processes and deliver more accu-
rate and responsive models. Furthermore, through scalability analysis and comparisons
with centralized approaches, TechSolve has highlighted BCFL’s superiority and fea-
sibility for real-world deployment. These findings underscore BCFL’s transformative
potential in addressing the evolving challenges of collaborative machine learning across
diverse industries. Looking ahead, TechSolve remains committed to advancing BCFL
methodologies, and driving continued innovation in collaborative machine learning
while upholding principles of privacy, security, and efficiency. The success of BCFL
at TechSolve not only showcases the power of blockchain technology in collaborative
learning but also paves the way for future advancements and applications in this domain.

Future Work
The future of blockchain-enhanced federated learning presents exciting avenues
for research and development, aiming to further enhance the collaborative learning
paradigm while addressing emerging challenges. One key area of focus is scalability
enhancement, with efforts directed toward optimizing systems to efficiently accommo-
date a larger number of participating nodes without compromising speed or security.
The development of dynamic privacy-preserving mechanisms is crucial, leveraging
blockchain for adaptable privacy models that can adjust based on data sensitivity and
user preferences. Interoperability and standardization efforts are needed to enable
seamless collaboration and data sharing across diverse blockchain networks and fed-
erated learning systems. Ethical considerations and regulatory frameworks require
in-depth exploration to ensure compliance with evolving privacy laws and ethical
guidelines. Real-world deployment and use cases across various industries will provide
valuable insights into the effectiveness and scalability of blockchain-enhanced feder-
ated learning. Designing incentive mechanisms within these ecosystems will encour-
age active participation while maintaining robust data privacy and security measures.
Continuous advancements in the robustness and security of blockchain-federated
learning systems will be paramount to mitigating potential vulnerabilities and adver-
sarial attacks. This evolving landscape holds significant promise for revolutionizing
collaborative learning while safeguarding data privacy, setting the stage for a more
resilient and widely applicable blockchain-enhanced federated learning framework.

REFERENCES
[1] Mishra, A., Garg, Y., Pandey, O. J., Shukla, M. K., Vasilakos, A. V., & Hegde, R. M.
(2024, July–August). A Novel Resource Management Framework for Blockchain-Based
Federated Learning in IoT Networks. IEEE Transactions on Sustainable Computing,
9(4), 648–660. doi: 10.1109/TSUSC.2024.3358915
110 Artificial Intelligence Using Federated Learning

[2] Tong, Z., Wang, J., Hou, X., Chen, J., Jiao, Z., & Liu, J. (2024). Blockchain-Based
Trustworthy and Efficient Hierarchical Federated Learning for UAV-Enabled IoT
Networks. IEEE Internet of Things Journal, 11(21), 34270–34282. doi: 10.1109/JIOT.
2024.3370964
[3] Zhang, H., Jiang, S., & Xuan, S. (2024). Decentralized Federated Learning Based on
Blockchain: Concepts, Framework, and Challenges. Computer Communications, 216,
140–150.
[4] Sadineni, G., Singh, J., Rani, S., Rao, G. S., Pasha, M. J., & Lavanya, A. (2024).
Blockchain-Enhanced Vehicular Ad-hoc Networks (B-VANETs): Decentralized Traffic
Coordination and Anonymized Communication. International Journal of Intelligent
Systems and Applications in Engineering, 12(1s), 443–456.
[5] Wang, P., Zhao, Y., Obaidat, M. S., Wei, Z., Qi, H., Lin, C., & Zhang, Q. (2022).
Blockchain-Enhanced Federated Learning Market with Social Internet of Things.
IEEE Journal on Selected Areas in Communications, 40(12), 3405–3421.
[6] Qu, Y., Gao, L., Luan, T. H., Xiang, Y., Yu, S., Li, B., & Zheng, G. (2020). Decentralized
Privacy Using Blockchain-Enabled Federated Learning in Fog Computing. IEEE
Internet of Things Journal, 7(6), 5171–5183.
[7] Miao, Q., Lin, H., Hu, J., & Wang, X. (2022). An Intelligent and Privacy-Enhanced
Data Sharing Strategy for Blockchain-Empowered Internet of Things. Digital
Communications and Networks, 8(5), 636–643.
[8] Xianjia, Y., Queralta, J. P., Heikkonen, J., & Westerlund, T. (2021). Federated learning
in Robotic and Autonomous Systems. Procedia Computer Science, 191, 135–142.
[9] Lv, P., Xie, L., Xu, J., Wu, X., & Li, T. (2022). Misbehavior Detection in Vehicular Ad
Hoc Networks Based on Privacy-Preserving Federated Learning and Blockchain. IEEE
Transactions on Network and Service Management, 19(4), 3936–3948.
[10] Firdaus, M., Larasati, H. T., & Rhee, K. H. (2022, June). A Secure Federated Learning
Framework Using Blockchain and Differential Privacy. In 2022 IEEE 9th International
Conference on Cyber Security and Cloud Computing (CSCloud)/2022 IEEE 8th
International Conference on Edge Computing and Scalable Cloud (EdgeCom) (pp.
18–23). IEEE.
[11] Liu, Y., Liu, P., Jing, W., & Song, H. H. (2023, December 15). PD2S: A Privacy-
Preserving Differentiated Data Sharing Scheme Based on Blockchain and Federated
Learning. IEEE Internet of Things Journal, 10(24), 21489–21501. doi: 10.1109/
JIOT.2023.3295763
[12] Huang, X., Han, L., Li, D., Xie, K., & Zhang, Y. (2023). A Reliable and Fair Federated
Learning Mechanism for Mobile Edge Computing. Computer Networks, 226, 109678.
[13] Shen, G., Fu, Z., Gui, Y., Susilo, W., & Zhang, M. (2023). Efficient and Privacy-
Preserving Online Diagnosis Scheme Based on Federated Learning in E-healthcare
System. Information Sciences, 119261.
[14] Fan, M., Yu, H., & Sun, G. (2021, November). Privacy-preserving Aggregation Scheme
for Blockchained Federated Learning in IoT. In 2021 International Conference on
UK-China Emerging Technologies (UCET) (pp. 129–132). IEEE.
[15] Billah, M., Mehedi, S. T., Anwar, A., Rahman, Z., & Islam, R. (2022). A Systematic
Literature Review on Blockchain Enabled Federated Learning Framework for Internet
of Vehicles. arXiv preprint arXiv:2203.05192.
[16] Ullah, I., Deng, X., Pei, X., Mushtaq, H., & Uzair, M. (2023). IoV-SFL: A Blockchain-
based Federated Learning Framework for Secure and Efficient Data Sharing in the
Internet of Vehicles. doi: 10.21203/rs.3.rs-3648280/v1
[17] Moudoud, H., & Cherkaoui, S. (2023). Multi-Tasking Federated Learning Meets
Blockchain to Foster Trust and Security in the Metaverse. Ad Hoc Networks, 150,
103264.
Blockchain-Enhanced Federated Learning 111

[18] Zhang, Y., & Yu, H. (2022). Towards Verifiable Federated Learning. arXiv preprint
arXiv:2202.08310.
[19] Li, W., Li, Y., Zheng, C., & He, R. (2023, August). Blockchain-based Model for Privacy-
enhanced Data Sharing. In 2023 10th International Conference on Dependable
Systems and Their Applications (DSA) (pp. 406–417). IEEE.
[20] Firdaus, M., Larasati, H. T., & Rhee, K. H. (2023). A Blockchain-Assisted Distributed
Edge Intelligence for Privacy-Preserving Vehicular Networks. Computers, Materials &
Continua, 76(3).
[21] Soltani, B., Haghighi, V., Mahmood, A., Sheng, Q. Z., & Yao, L. (2022, October).
A Survey on Participant Selection for Federated Learning in Mobile Networks.
In Proceedings of the 17th ACM Workshop on Mobility in the Evolving Internet
Architecture (pp. 19–24). Association for Computing Machinery.
[22] Fan, M., Ji, K., Zhang, Z., Yu, H., & Sun, G. (2023, September 15). Lightweight Privacy
and Security Computing for Blockchained Federated Learning in IoT. IEEE Internet of
Things Journal, 10(18), 16048–16060. doi: 10.1109/JIOT.2023.3267112
[23] Otoum, S., Al Ridhawi, I., & Mouftah, H. (2021). Securing Critical IoT Infrastructures
with Blockchain-Supported Federated Learning. IEEE Internet of Things Journal,
9(4), 2592–2601.
[24] Huang, C., & Liu, S. (2024). Securing the Future of Industrial Operations: A Blockchain-
Enhanced Trust Mechanism for Digital Twins in the Industrial Internet of Things.
International Journal of Computers and Applications, 1–10.
[25] Wang, L., & Guan, C. (2024). Improving Security in the Internet of Vehicles:
A Blockchain-Based Data Sharing Scheme. Electronics, 13(4), 714.
[26] Wan, T., Jiang, T., Liao, W., & Jiang, N. (2024). Hierarchical Incentive Mechanism
for Federated Learning: A Single Contract to Dual Contract Approach for Smart
Industries. International Journal of Intelligent Systems, 2024.
[27] Aljabri, A., Jemili, F., & Korbaa, O. (2024). Convolutional Neural Network for Intrusion
Detection Using Blockchain Technology. International Journal of Computers and
Applications, 46(2), 67–77.
[28] An, D. A. G. (2024, January). Jiakai Hao¹, Guanghuai Zhao¹, Ming Jin¹, Yitao Xiao²),
Yuting Li¹, and Jiewei Chen². In Proceedings of the 13th International Conference on
Computer Engineering and Networks: Volume III (Vol. 1127, p. 281). Springer Nature.
[29] Selvakumar, V., Maaliw, R. R., Sharma, R. M., Oak, R., Singh, P. P., & Kumar, A.
(2024). Blockchain-Aware Federated Anomaly Detection Scheme for Multivariate
Data. In Artificial Intelligence, Blockchain, Computing and Security Volume 1
(pp. 690–698). CRC Press.
 7 Federated Learning-Based
Smart Transportation
Solutions
Deploying Lightweight
Models on Edge Devices
in the Internet of Vehicles
Sivabalan Settu, Raveendra Reddy,
Appalaraju Muralidhar, Thangavel
Murugan, and Rathipriya Ramalingam

INTRODUCTION
The Internet of Things (IoT) and the proliferation of linked devices have advanced
quickly, revolutionizing several industries, including transportation. An extension of
the Internet of Things, the IoV paradigm has become a disruptive force that allows
data interchange and seamless connection between automobiles, roadside units, and
cloud infrastructure. Current generation focuses are intelligent transportation sys-
tems that can optimize traffic flow, increase road safety, and boost overall efficiency.
However, there are a lot of issues with data privacy, bandwidth limitations, and pro-
cessing complexity because of the vast volume of data that IoV devices generate.
Conventional, centralized methods of data processing and machine learning
model training might not be desirable or feasible in such dispersed and resource-
constrained scenarios. Without requiring raw data exchange, FL is a collaborative
machine-learning technique that allows training models across numerous dispersed
devices. With FL, devices can train local models on their data and simply share the
model updates with a central server or aggregator, as opposed to centralizing all of
the data on a single server. Here, multi-input and multi-output signal propagation
methods are used, the computing effort is split among several devices, communica-
tion overhead is decreased, and data privacy is maintained [1–5].
Deploying lightweight models on edge devices, including RSUs and OBUs, is criti-
cal in the context of the IoV. Because of their frequently constrained computational
capabilities, these edge devices make it difficult to implement sophisticated machine-
learning models. For edge device deployment, lightweight models provide a workable

112 DOI: 10.1201/9781003482000-7

Federated Learning-based Smart Transportation Solutions 113

option by balancing model performance and computing needs. This study investigates
the use of application delivery networks for edge devices in conjunction with federated
learning for smart transportation solutions inside the IoV ecosystem. Developers hope
to overcome the issues of data privacy, bandwidth limitations, and computational limi-
tations by utilizing the advantages of these two complementary approaches, allowing
for more intelligent decision-making, more efficiency, and improved safety.
The study explores the foundational ideas of application delivery network and fed-
erated learning, emphasizing the benefits and drawbacks of each. It delves deeper into
how these technologies fit into the IoV ecosystem, looking at a range of applications
such as collaborative sensing, intelligent vehicular assistance, predictive maintenance,
and traffic flow management. The goal is to show that deploying lightweight models
based on FL on edge devices in the IoV is both feasible and effective, using theoretical
analysis, simulations, and real-world case studies. This section also points out pos-
sible drawbacks, difficulties, and future lines of inquiry to develop the field and open
the door for the general use of these technologies in smart transportation solutions.
The present text solves the significant issues regarding information privacy, lack
of resources, and computer processing boundaries in an attempt to assist in enhanc-
ing safer, more ecologically sound, and energy-effective modes of transport, which
in turn advance the entire standard that life brings and promote a transition regard-
ing increasingly intelligent and associated towns and villages. An example scenario
mentioned in Figure 7.1 a block of social networks, numerous vehicles and links, and
cognitive and auto-taxi. Table 7.1 is the symbol and abbreviation used in this chapter.

FIGURE 7.1 Vehicle to everything connection establishment in a different vehicle

114 Artificial Intelligence Using Federated Learning

TABLE 7.1
Federated learning based Internet of Vehicles state and
their abbreviation
State Abbreviation

Internet of Vehicle IoV

Internet of Things IoT
Federated Learning FL
Machine Learning ML
Roadside Assistance Devices RSUs or RSDs
On-Board Gadgets OBUs
Edge Computing EC
Driver Assistance Technologies DAT
Vehicle Data Distribution VDD
Horizontal Federated Learning HFL
Vertical Federated Learnings VFL
Application Delivery Network AND
Vehicle Data VD
Global Server GS
Client Vehicle CV
Blockchain BC
Application Binary Interface ABI
Smart Transportation Contract STC

BACKGROUND STUDY
IoV Paradigm
The IoV notion is a game-changer for the transportation industry. It uses cloud com-
puting facilities, curbside measurements, and vehicle linkages to allow for cognitive
choices, smooth interaction, and knowledge trade. The new model can completely
reshape public transit via enhanced reliability, effectiveness, and safeguarding. It is
an outgrowth of the broader IoT idea. A vast system of interdependent components,
such as automotive vehicles with OBUs, RSUs, and platforms built on the cloud,
appear in the IoV ecosystem. All of these components acquire and transmit data on
a regular schedule about a wide range of areas, including surroundings, circulation
patterns, driving actions, road hazards, and vehicle telemetry systems. For the intent
of public transportation optimization and the growth of a multitude of innovative
instances, this data can be highly significant. Figure 7.2a, and Figure 7.2b dis-
play the Inner, Outer link establishments in ADN at numerous road-type ground
transportation.
Federated Learning-based Smart Transportation Solutions 115

FIGURE 7.2a Outside sensor establishments

FIGURE 7.2b Inside sensor establishments for autonomous

116 Artificial Intelligence Using Federated Learning

Challenges in the IoV Ecosystem

The IoV paradigm offers many opportunities, but it also brings with it several
obstacles that need to be tackled to fully capitalize on its true potential. The
security and privacy of data are two of the key problems. The critical character
of the data gathered through transport vehicles and building components gives
reasons for suspects to be concerned about breaches of privacy, prohibited entry
points, and wrongful use of sensitive information. The limitations on access and
communication costs encountered with moving enormous volumes of data from
various handheld devices onto central machines or cloud-based systems provide
a further significant obstacle. In such decentralized and constrained resource sce-
narios, classically consolidated techniques to handle data and model development
for machine learning could prove really wanted or realistic. Similarly, the small
processing capacity of edge devices, including OBUs and RSUs, frequently inhib-
its the implantation of more sophisticated machine learning models on each other.
For the sake of intelligent choices at the edge, these capacities limit the authority
of the building as well as the adoption of lightweight models intended exclusively
for border equipment.

Federated Learning: A Centralized/Decentralized Approach

In the IoV ecosystem, FL has shown promise as a decentralized machine learn-
ing method for addressing issues with data privacy, bandwidth limits, and process-
ing constraints. Without requiring the sharing of raw data, FL allows collaborative
training of machine learning models across numerous decentralized devices. Under
the FL paradigm, every device or party uses its data to train a local model; only
the model updates, or gradients, are shared with an aggregator or central server.
Sensitive data is kept localized in this method to protect data privacy, and only
model updates—which are usually much smaller than raw data—are transmitted
to minimize communication costs. FL has two types called horizontal, and vertical
tabulated in Figure 7.3.

Lightweight Models for Edge Devices

The setting up of lightweight models on limited-resource edge devices is criti-
cal to enabling smart choices at the edge and utilizing the benefits of FL in the
IoV ecosystem. Lightweight models can be distributed on edge devices with lim-
ited capabilities because the choices are designed to reconcile the performance of
models and computation needs. Model enlargement, quantization, and reduction
are some of the techniques that can be used to either enhance contemporary deep-
learning models or create new lightweight architectures designed specifically for
IoV workloads. Using the FL procedure, these lightweight models may be coop-
eratively trained to achieve the greatest utilization of resources and retain privacy
based on separated knowledge. To address the issues of data privacy, bandwidth
limitations, and computational aspect boundaries, academics and professionals in
Federated Learning-based Smart Transportation Solutions 117

FIGURE 7.3 Types of Federated Learning

the IoV domain anticipate integrating FL and lightweight models. This will enable
skilled modes of transport that improve safety, efficiency, and long-term viability
in public transit systems. Finally, Figure 7.4 shows the section study of this book
chapter.
118 Artificial Intelligence Using Federated Learning

FIGURE 7.4 Organization of the book chapter sections

Federated Learning-based Smart Transportation Solutions 119

FIGURE 7.5 Federated Edge Devices Learning and study of central, dispersed system

End Background Study

Finally, at the end of the section study, the FL type was used to centralize and
disperse server data transactions, and applicable ground features were identi-
fied. Figure 7.5 presents the federated learning from vehicle edge devices and
the possibilities of feature enhancement in on-load, off-load, and live 6G vision
directions.

PROPOSED WORK
The IoV paradigm has brought about an immense shift in the world of transport in
the era of the IoT. IoV makes use of the connectivity between cars, roadside equip-
ment, and cloud infrastructure to facilitate intelligent decision-making, simple inter-
action, and information transfer. Still, challenges with the confidentiality of data,
shortages of bandwidth, and information technology inefficiency originate from
the tremendous quantity of data generated through IoV devices. A flexible machine
learning technology called FL is now coming to light as an acceptable means to get
past such barriers as shown in Figure 7.6.
120 Artificial Intelligence Using Federated Learning

FIGURE 7.6 Federated Learning in the IoV ecosystem, autonomous flow

Deploying Lightweight Models on Edge Devices

Complex machine learning models can be computationally and highly energy-intensive
to deploy when it comes to devices that are limited in resources. To overcome this bar-
rier, specialists propose creating and implementing lightweight models made especially
for edge devices throughout the IoV. The construction of lightweight models intends to
combine processing constraints with the performance of the model. Models like these
can be distributed on device edges with limited capabilities because they usually con-
sume fewer settings and a smaller amount of power. Experts could develop new light-
weight topologies designed specifically for Internet of Vehicles applications or perfect
established deep learning models by applying methods like model shrinkage, quantiza-
tion, and cutting down. An example scenario of the dynamic on-load/offload mechanism
of FL is stated in Figure 7.4. The diagram represents model deployment using agents
that link numerous things like inside edge computes, including road, rail, air, and water
transportation systems. It includes a model_edge (zone and region), a model_vehicle (in
and out links), and a model_RSU (RSU1,. . ., RSU n). Figure 7.7 spots the vehicle at a
time of movement used to find the social affinity application destination.

Applications of FL in Smart Transportation

FL has various advantages in the field of technological transportation when it inte-
grates with lightweight models that are stationed at the forefront of devices.
Federated Learning-based Smart Transportation Solutions 121

FIGURE 7.7 IoV Region-based dynamic offload/on-load edge devices for inter and
intra-communication

Traffic Stream Optimization: The intelligent routing method and regulation of

signals are used by FL models to anticipate congestion, recognize slowdowns, and
improve the efficiency of traffic by analyzing instantaneous information from cars
and roadside sensors for analysis.
Automated Transportation Assist: By utilizing data from diverse automobiles
and road sections, FL models can enable refined Driver Assistance Technologies
(DAT) and autonomous driving abilities such as people-walking recognition, steer-
ing avoidance, and lane-keeping aid.
Automatic Maintenance: While data on highway circumstances, attitude of
drivers, and car performance can be gathered via IoV devices, through the analysis
of these data points, FL models can increase vehicle safety and effectiveness by
modeling service needs, optimizing restoration agendas, and mitigating mistakes.
Ecological Inspection: Information on emissions, air quality, and noise levels
can be gathered using IoV devices. By analyzing this data, FL models can identify
polluting tourist attractions, improve the speed of traffic, and create policies to lessen
the adverse impact travel has on the ecosystem.
Collaborating Sensed: FL models can facilitate collaborating perceiving apps,
like joint perceiving for greater awareness of the situation and imminent risk identi-
fication, by utilizing data from many vehicles and checkpoint gadgets.
122 Artificial Intelligence Using Federated Learning

Challenges and Future Directions

Given that cooperative learning and lightweight models are currently being devel-
oped as appealing options for smart mobility in the IoV, significant barriers remain:
Successful modes of communication and cost optimization strategies have to be
used for handling updated models along with data spread via edge devices and the
primary server or aggregators.
Protection and Safety: Solid security-preserving approaches and safe aggregate
procedures are critical for protecting data privacy and mitigating aggressive attacks
or model poisoning.
Divergence and Ability to Scale: IoV devices possess various hardware abilities,
data distributions, and other relationship boundaries. FL algorithms must be formed
to address variety while expanding to accommodate plenty of devices.
Models Customization: Though FL enables shared learning, it is a requirement
for geared models that address special device or client demands, such as specific
driving choices or local circumstances.
Uniformity and Coordination: For broad adoption and accessibility, the FL
solution must seamlessly fit into the present infrastructure for transportation, as well
as technologies and endpoints that are universal.
As the IoV ecology advances, unified learning, and lightweight models will prove
key in allowing knowledgeable transportation solutions to utilize dispersed informa-
tion while protecting the safety and promoting the utilization of resources. Further
study and partnership among academia, industry, and government agencies will be
important in solving barriers and attaining the entire potential of these advances in
smart transportation [6–11, 12, 13].

FL DATA PROCESS STEPS VIA APPLICATION DELIVERY

NETWORK (ADN) AND DATA ANALYTICS
FL is a modular machine learning technology that allows for joint model training
across numerous devices or entities without the need for centralization or substitu-
tion of data in its completeness. The following Figure 7.8 illustrates the data pro-
cessing steps involved in ADN.
This paradigm gained substantial acceptance in the world of statistical analysis,
particularly in circumstances in which confidentiality of data, capacity restrictions,
and computation restraints are noteworthy. According to the FL-IoV use of federated
data analytics proves essential in that it allows competent modes of transportation
while focusing on the barriers posed by the IoV ecological systems.

Federated Studies and Analytics for Data Throughout the IoV

The IoV structure consists of a massive chain of coupled transport vehicles, outside
gear, and the cloud that achieve massive volumes of knowledge. This data comprises
car telemetry, which includes driver actions, circulation patterns, roadway circum-
stances, and aspects of the environment. Federated learning data analytics lets you
perform the evaluation and representation of scattered information without the need
to organize it, hence safeguarding the confidentiality of information and eliminating
Federated Learning-based Smart Transportation Solutions 123

FIGURE 7.8 ADN typical steps involved in data processing

communication overhead [14]. When deploying lightweight models on edge devices in

the IoV ecosystem, edge devices such as OBUs and RSUs have limited computational
resources. Deploying complex machine learning models on these resource-constrained
devices can be computationally intensive and energy-inefficient. Federated learning
data analytics addresses this challenge by enabling the training and deployment of
lightweight models tailored specifically for edge devices. Lightweight models are
designed to strike a balance between model performance and computational require-
ments. These models are typically smaller in size, have fewer parameters, and require
less computational power, making them suitable for deployment on edge devices with
limited resources. Federated learning data analytics facilitates the collaborative train-
ing of these lightweight models across multiple edge devices, leveraging the distrib-
uted data while preserving privacy and reducing communication overhead. Figure 7.4
shows the region IoV getting the FL training experience [15, 16].

Key Aspects of FL Data Analytics

Lightweight and comfortable: models aim to strike an appropriate balance between
model efficacy and computation demands. These kinds of models are generally less
extensive in scope, have fewer features, and require less computational dominance,
124 Artificial Intelligence Using Federated Learning

FIGURE 7.9 Edge devices with FL approach among RSU, OBU, Infrastructure in cell

making them optimal for use on edge devices with a limited number of resources.
FL statistical analysis allows for collective training of such tiny machines span-
ning plenty of edge technology devices using disparate information while preserving
privacy and eliminating expenses for communication. The fourth figure illustration
exhibits the region IoV gaining FL training experiences. Figure 7.9 displays the cell
within a cell or cell-free technology where infrastructure and RSUs are bonding with
the corresponding network.
Basic Benefits of FL Data Analytics, Modular Learning: Federated learning
data aggregation allows for the dispersed training of statistical ensembles across
various devices or entities. Each device or party builds its model using its data, and
only updated models or variations are transferred to an underlying server or aggre-
gator. The confidentiality of information is safeguarded by federated learning data
analytics, which resists gathering raw data. Specific gadgets or persons maintain
oversight of their secret information, avoiding safety and legal concerns.
Spectrum Efficacy: Compared to passing huge quantities of raw data, federated
learning statistical analysis mandates only the distribution of variations in models
or shadings, which are frequently much smaller in size. It features reduced transfer
costs and bandwidth requirements that make it ideal for resource-constrained IoV
Federated Learning-based Smart Transportation Solutions 125

TABLE 7.2
FL- IoV using Smart Transport using ADN Infrastructure, Coverage, Road/Lane,
Cell or Cell-free
Procedure 1: Smart Transportation Contract (STC)
{Vehicle (V): Road, Rail, Water, Air, Space . . .} Via Application Delivery Network
(ADN1, ADN2, . . ., ADNn)
Vehicle Registration {V1, V2, . . ., Vn} // Number of Vehicle Data (VD),
if {V1, V2, . . ., Vn} == successful then // Global Server (GS)
Check {VD 1,. . . . VD n} from the GS // Client Vehicle (CV)
else
Register {CV1, . . . CV n} as a New {CV}
U ← CV
Store Data on BC // Blockchain (BC)
Encrypt data Using Sha256 Algo
Result generate ABI of Contract // Application Binary Interface (ABI)
generate Byte code of Contract
Decrypt data Using Sha256 Algorithm
end if

applications. Table 7.2 gives the smart transport mechanism procedures and the
numerous application delivery network points.
Regional Modeling: FL Informatics is capable of managing diverse data
deployments from several devices or entities. Each regional model undergoes
training on input from the relevant faction, and the procedure of aggregation com-
bines all the different viewpoints and generates more reliable as well as generic
forecasts [17, 18].
Continuous Learning: As newly collected information is obtainable at specific
devices or those involved, FL data analytics offers continuous learning and update
services, making sure deploying systems stay relevant and versatile to evolve with
scenarios. Application for Intelligent Mobility FL analytics of data has numer-
ous uses in smart mobility and the IoV ecosystem, including congestion oversight,
cognitive vehicular assistance, automatic upkeep, surveillance of the environment,
and joint sense. FL data analytics allows for more sophisticated decision-making,
enhanced security, and improved productivity in public transport by using dispersed
information while safeguarding data and enhancing resource utilization.

RESULT STUDY
The joint use of FL and lightweight models distributed on edge equipment in the
IoV ecosystem has shown favorable outcomes in facilitating intelligent modes of
transport while tackling the confidentiality of data, access limits, and computational
aspects of barriers. Figures 7.10 and 7.11 represent the overall diagram of smart
transportation in the city.
126 Artificial Intelligence Using Federated Learning

FIGURE 7.10 Proposed Systematic diagram for Fed-IoV

Federated Learning-based Smart Transportation Solutions 127

FIGURE 7.11 Federated Learning-based Smart Transportation

Safeguarding Privacy and Collaborative Learning

The distributed architecture of FL promotes collaborative learning without the need
to disclose unprocessed data, hence safeguarding the confidentiality of data in the
IoV ecosystem. Specific devices or entities maintain oversight of their proprietary
information by training individual models on their knowledge and only exchang-
ing modifications to the models with global aggregators. This approach responds
to safety and legal issues, both of which are particularly significant in the vehicle
industry, as intimate and automotive data are associated.

Reduced Communication Overhead

One of FL’s core advantages is its capacity to diminish transmission costs and band-
width demands. Instead of sending huge amounts of data in raw form from multiple-
state IoV devices to centrally located servers, FL just needs the sharing of changes
to models or shadings, which are often significantly smaller in size. This method is
especially useful in the IoV ecosystem, where access limits and large data quantities
offer notable obstacles.

Efficient Edge Computing

The setup of lightweight models on hardware such as OBUs and RSUs gives rapid
computation and intelligent choices to be made. Lightweight models pound the right
balance between space and practical demands by utilizing tackles such as model
expansion, the goal of quantification, and sprucing which make them acceptable for
use on constrained gateway workstations.
128 Artificial Intelligence Using Federated Learning

Multidisciplinary Processing of Data

FL can handle various data disseminations across several IoV systems or individu-
als. Each local model is trained on data from the associated party, and the aggrega-
tion process combines all of these viewpoints to create more robust and universal
models. This feature is especially important in the IoV ecological system, where
disparity in information is essential due to varied vehicle categories, driving designs,
and surrounding factors.

Continuous Growth and Adaptability

As the latest information becomes accessible from specific IoV devices or third
parties, FL sells continuous adaptation and modeling upgrades, making sure the
set-up models are up to date-and flexible with evolving conditions. Such func-
tionality comes in handy in a dynamic journey scenario where commute pat-
terns, surface conditions, and the behavior of travelers can change promptly. If
you consider that the social affinity application and delivery network depend on
mobility, it shows the vehicle driver necessary things like a fuel station, restau-
rant, temple, etc.
Finally, future directions are illustrated in Figure 7.12, and the natural envi-
ronment protection and ecology system is suitable for handling social delivery
networks.
Real-world approaches and case investigations The findings of the research
show that employing FL-based lightweight models on edge technology is practical
and efficient in a variety of IoV usages, including the flow of traffic optimization,
smart vehicular guidance, proactive upkeep, environmental tracking, and mutually
beneficial sensing. The suggested approach brought about favorable outcomes in
terms of smoother circulation of traffic, increased safety on the roadways, efficient
use of assets, and reduced greenhouse gas emissions, as indicated by simulations
and case studies. These outcomes line up with the overall objectives of intelligent
transit systems, resulting in improved effectiveness, profitability, and safer mobil-
ity choices.

FIGURE 7.12 Future Direction enhancement

Federated Learning-based Smart Transportation Solutions 129

Addressing Challenges and Future Directions

While incorporating FL and compact designs leads to promising outcomes, various
difficulties, and limitations have been acknowledged, paving the way for future stud-
ies in these domains.
Network Efficiency: Quick protocols for communicating and spectrum optimi-
zation techniques are needed to support new models and data transmission between
devices at the edge and the data aggregator, therefore eliminating communication
waste.
Volatility and ability to scale: As the IoV ecosystem expands, FL solutions and
lightweight concept structures must be designed to deal with rising variability and
scale to handle numerous devices with diverse characteristics.
Enhancements to privacy and security: The connection followed strong
privacy-protection tactics and secure collection, as well as well-known protocols.
Those are essential to ensuring data confidentiality and preventing provocative attacks
or paradigm poisoning, especially in security-sensitive applications for transport.
Model Customization: While FL encourages knowledge sharing from tiny to
large edges, there is a need for models tailored to individual devices or user needs,
such as riding preferences or local environments.
Uniformity and Integration: Easy integration of FL solutions into existing pri-
vate, public, and industrial transport systems, as well as the creation of common
interfaces and screens, is critical for broad adoption and connectivity among distinct
IoV components and stakeholders.

RESULT, CONCLUSION, AND STUDY

The simultaneous use of FL and lightweight designs set up on edge devices in the
IoV ecosystem exhibited promising results in allowing intelligent mobility solutions
while addressing safeguarding information, access limits, and processing concerns.
This study reviewed the viability and efficacy of the suggested approach by using
comprehensive exercises, case studies from real life, and theoretical analyses. The
main results and inferences are discussed as follows:
Freedom-Preserving Joint Knowledge Acquisition: By making use of FL’s
modular structure, the approach recommended allows for teamwork without expos-
ing raw data, accordingly safeguarding the confidentiality of data in the IoV indus-
try. This strategy overcomes concerns regarding privacy as well as governmental
hardships, both of which are vital in the transport industry.
Decreased relations expense: Whenever placing FL-based scenarios, updated
models or differential transmission must be performed, which causes substantially
fewer communication expenses and demands on bandwidth than typical consoli-
dated techniques.
Rapid Boundary Computations: The use of portable designs specially designed
for constrained-in-resource edge devices like OBUs and RSUs promotes economic
cloud computing and smart choice-making at the edge.
Dealing with disparate signals: FL’s ability to manage diverse data distributions
across different IoV devices and organizations has culminated in more robust and
130 Artificial Intelligence Using Federated Learning

FIGURE 7.13 FL is categorized by a centralized, distributed system

overall models that account for an array of driving sequences, vehicle formats, and
circumstances in the environment.
This is the final section of the chapter, and Figure 7.13. indicates the FL type with
central and dispersed server systems. Figure 7.14 shows the sample network settings
defined and the numerous parameters used. These settings are used to continuously
acquire and update ground awareness about themes like traffic, road damage, politi-
cal meetings, etc. Especially take one social application to check individual vehicle
identity, range, current ground city updates, cloud database retrieval, on-load time,
off-load time, and live.
Continuous Acquisition of Knowledge and Ground Alterations: This approach
supports ongoing learning and revisions to models, ensuring that positioned models
stay current and responsive to shifting circumstances in a dynamic traffic ecosystem.
This traffic crowd recognition technology is applied to guarantee a safe path.
Real-World Use Cases: A research study found that the FL technique is appli-
cable and effective in an assortment of IoV functions, such as streamlining traffic
optimization, intelligent vehicle advice, automatic upkeep, environmental surveil-
lance, and joint senses.
Greater efficiency of transportation and sustainable society: This study’s con-
clusion helps foster the creation of effective, less hazardous, and green public transit
Federated Learning-based Smart Transportation Solutions 131

FIGURE 7.14 The setting of distributed CV selection

systems by overcoming issues related to the privacy of data, optimizing the usage of
resources, and enabling automated decision-making.
While the proposed solution has yielded promising results, several challenges
and limitations have been identified, presenting opportunities for future research.
These include communication efficiency, handling increasing levels of heterogeneity
and scalability, enhancing privacy and security mechanisms, enabling model per-
sonalization, and facilitating seamless integration and standardization across the IoV
ecosystem. By addressing these challenges and leveraging the synergies between
FL and lightweight models, the research community and industry stakeholders can
unlock the full potential of intelligent transportation solutions in the IoV ecosystem,
contributing to more efficient, secure, and sustainable mobility systems. Continued
research, collaboration, and innovation in this domain are crucial for realizing the
vision of smarter, safer, and more connected transportation networks.

CONCLUSION
The deep study begun in this study explores the pairing of FL with compact
designs set up on interfaces within the context of IoV ecosystems. It launched
major discoveries with insights from earlier studies, conceptual frameworks, and
132 Artificial Intelligence Using Federated Learning

observational data, resulting in an in-depth grasp of the strategy’s capabilities and

imperfections. The meta-analysis noticed that combining FL and portable archi-
tectures provides an approach to a variety of barriers in the IoV natural context,
including concerns about safeguarding information, shortages of resources, and
edge device estimation boundaries. By using FL’s global structure and resource-
efficient architecture of flexible entities, by capturing FL’s decentralized feature
and the energy-efficient design of ultralight models, this method offers cognitive
solutions for transportation that prioritize data privacy, maximize utilization of
resources, and facilitate smart choices at the frontier. It responded to FL’s funda-
mental principles, like its ability to aid cooperation in learning by encapsulating
initial data, safeguarding sensitive facts, and bringing down link costs. In addition,
the scientific review assessed distinct methods for utilizing featherweight circum-
stances such as simulation enlargement and quantization. This process and savings
are appropriate for distribution on limited-resource peripherals. Since emerging
optimistic, it has discovered significant odds such as relationship success and data
safeguards, which demand precision. It also highlighted future research opportu-
nities for model-based personalized support and easy entry into the IoV platform.
It provides a solid foundation for the future rise of FL-based technologies with
handheld designs in the IoV sector.

REFERENCES
1. Samarakoon, S., Bennis, M., Saad, W., & Debbah, M. (2019). Federated learning for
ultra-reliable low-latency V2V communications. In 2019 IEEE Global Communications
Conference (GLOBECOM) (pp. 1–7). IEEE.
2. Lu, Y., Huang, X., Zhang, K., Maharjan, S., & Zhang, Y. (2021, January/February).
Blockchain and federated learning for 5G beyond. IEEE Network, 35(1), 219–225.
http://doi.org/10.1109/MNET.011.1900598.
3. Lim, W. Y. B., Nguyen Cong Luong, Dinh Thai Hoang, Yutao Jiao, Ying-Chang Liang,
Qiang Yang. (2020). Federated learning in mobile edge networks: A comprehensive
survey. IEEE Communications Surveys & Tutorials, 22(3), 2031–2063. http://doi.
org/10.1109/COMST.2020.2986024.
4. Niknam, S., Dhillon, H. S., & Reed, J. H. (2019). Federated learning for wireless
communications: Motivation, opportunities, and challenges. IEEE Communications
Magazine, 58, 46–51.
5. Brecko, A., Kajati, E., Koziorek, J., & Zolotova, I. (2022). Federated learning for edge
computing: A survey. Applied Sciences, 12(18), 9124. https://doi.org/10.3390/app12189124.
6. Hinton, G., Vinyals, O., & Dean, J. (2015). Distilling the knowledge in a neural net-
work. arXiv preprint arXiv:1503.02531.
7. Wang, K., Liu, Z., Lin, Y., Lin, J., & Han, S. (2019). HAQ: Hardware-aware auto-
mated quantization with mixed precision. In Proceedings of the IEEE/CVF Conference
on Computer Vision and Pattern Recognition (pp. 8612–8620). IEEE. doi: 10.1109/
CVPR.2019.00881.
8. Jeong, E., Oh, S., Kim, H., Park, J., Bennis, M., & Kim, S. L. (2018). Communication-
efficient on-device machine learning: Federated distillation and augmentation under
non-iid private data. arXiv preprint arXiv:1811.11479.
9. Macedo, D., Santos, D., Perkusich, A., & Valadares, D. C. G. (2023). Mobility-aware
federated learning considering multiple networks. Sensors (Basel), 23(14), 6286. http://
doi.org/10.3390/s23146286. PMID: 37514581; PMCID: PMC10386473.
Federated Learning-based Smart Transportation Solutions 133

10. Sivabalan, S., Dhamodharavadhani, S., & Rathipriya, R. (2020). Arbitrary walk with
minimum length based route identification scheme in graph structure for opportunistic
wireless sensor network. In Aboul Ella Hassanien & Ashraf Darwish (Eds.), Intelligent
Data-Centric Systems, Swarm Intelligence for Resource Management in Internet of
Things (pp. 47–63). Academic Press. ISBN 9780128182871. https://doi.org/10.1016/
B978-0-12-818287-1.00006-1.
11. Kulanthaiyappan, S., Settu, S., & Chellaih, C. (2020). Internet of vehicle: Effects of tar-
get tracking cluster routing in vehicle network. In 2020 6th International Conference
on Advanced Computing and Communication Systems (ICACCS) (pp. 951–956).
Coimbatore, India. http://doi.org/10.1109/ICACCS48705.2020.9074454.
12. Sivabalan, S., & Rathipriya, R. (2017). Slot scheduling Mac using energy efficiency in ad
hoc wireless networks. In 2017 International Conference on Inventive Communication
and Computational Technologies (ICICCT) (pp. 430–434). Coimbatore, India. http://
doi.org/10.1109/ICICCT.2017.7975234.
13. Shao, Z., Yang, H., Xiao, L., Su, W., & Xiong, Z. (2023). Energy and latency-aware
resource management for UAV-assisted mobile edge computing against jamming. In
GLOBECOM 2023–2023 IEEE Global Communications Conference (pp. 1848–1853).
Kuala Lumpur, Malaysia. http://doi.org/10.1109/GLOBECOM54140.2023.10437090.
14. Sivabalan, S., & Rathipriya, R. (2023). Efficient energy resource selection in home
area sensor networks using non swarm intelligence based discrete venus flytrap search
optimization algorithm. Wireless Personal Communications, 128, 249–265. https://doi.
org/10.1007/s11277-022-09953-y.
15. Liu, S., Yang, H., Xiao, L., Zheng, M., Lu, H., & Xiong, Z. (2024). Learning-based
resource management optimization for UAV-assisted MEC against jamming. IEEE
Transactions on Communications. http://doi.org/10.1109/TCOMM.2024.3374356.
16. Kain, R., & Sorour, S. (2022). Worker resource characterization under dynamic usage
in multi-access edge computing. In 2022 International Wireless Communications
and Mobile Computing (IWCMC) (pp. 1070–1075). Dubrovnik, Croatia. http://doi.
org/10.1109/IWCMC55113.2022.9824299.
17. Gao, Z., Yang, L., & Dai, Y. (2023, June 1). Large-scale computation offloading using
a multi-agent reinforcement learning in heterogeneous multi-access edge computing.
IEEE Transactions on Mobile Computing, 22(6), 3425–3443. http://doi.org/10.1109/
TMC.2022.3141080.
18. Santos, Á., Bernardino, J., & Correia, N. (2023). Automated application deployment
on multi-access edge computing: A survey. IEEE Access, 11, 89393–89408. http://doi.
org/10.1109/ACCESS.2023.3307023.
 8 Application of Artificial
Intelligence and
Federated Learning in
Petroleum Processing
Abdelaziz El-hoshoudy

Abbreviations

0t is the global model parameter at time t.

n is the learning rate.
L (0t ; Dt ) is the loss function computed on the local data set (Dt) at time (t).
VL (0t ; Dt ) is the gradient of the loss function concerning the global model parameters.
i
0t+1 is the updated local model parameters at a time (t + 1) for the ith data source.
Dti is the local data set at the ith data source at a time (t).
N is the total number of participating data sources.
0t+1 is the aggregated global model parameters at time (t + 1).

INTRODUCTION
The petroleum industry, encompassing the exploration, extraction, refining, trans-
portation, and marketing of petroleum products, is a complex and critical sector for
the global economy. The petroleum industry, as a cornerstone of the global economy,
involves a series of intricate and interrelated processes, from the exploration of crude
oil to its final delivery as a variety of petroleum products (Singh, 2023). The integra-
tion of artificial intelligence (AI) and federated learning (FL) into this sector repre-
sents a paradigm shift, offering novel pathways to address longstanding challenges
related to efficiency, safety, and environmental sustainability. This exploration of
AI and FL’s application within petroleum processing elucidates their transformative
potential, focusing on operational optimization, predictive maintenance, and carbon
footprint reduction (Yussuf & Asfour, 2024). AI facilitates the optimization of vari-
ous petroleum industry operations through sophisticated algorithms that can analyze
vast data sets to predict outcomes, automate decision-making, and enhance the effi-
ciency of processes (Koroteev & Tekic, 2021). In exploration, AI algorithms interpret
seismic data to predict the likelihood of oil or gas presence, significantly reducing

134 DOI: 10.1201/9781003482000-8

Application of AI and FL in Petroleum Processing 135

the costs and environmental impact associated with drilling exploratory wells. In
refining, AI models optimize process parameters such as temperature and pressure
in real time, ensuring the optimal conversion of crude oil into high-value products
(Khaldi et al., 2023). One of the standout contributions of AI in this sector is the
advancement of predictive maintenance. By leveraging data collected from sensors
installed on equipment, AI models can predict failures before they occur, schedul-
ing maintenance only when necessary. This not only extends the life of equipment
but also minimizes downtime, thereby enhancing operational efficiency and safety.
Predictive maintenance exemplifies how AI can lead to substantial cost savings and
risk reduction in petroleum processing. Federated learning emerges as a revolution-
ary approach by enabling multiple stakeholders to collaboratively improve AI mod-
els without directly sharing sensitive or proprietary data (Rauniyar et al., 2024). In
the context of the petroleum industry, FL allows for the collective improvement of
models used in exploration, production, and refining, while ensuring compliance
with stringent data privacy regulations and corporate confidentiality policies. FL
facilitates the deployment of AI models that benefit from diverse data sources across
different geographic locations and operational environments. This is particularly
beneficial for real-time monitoring and anomaly detection, where models trained on
a wider array of data can more accurately identify potential issues, from equipment
malfunctions to unsafe operational conditions, thereby enhancing safety and reduc-
ing environmental risks.
Both AI and FL contribute to the petroleum industry’s efforts to reduce its carbon
footprint. By optimizing operations, these technologies reduce the energy consump-
tion and greenhouse gas emissions associated with petroleum processing. Moreover,
AI and FL can aid in carbon capture and storage (CCS) technologies, improving
their efficiency and enabling more sustainable petroleum processing practices.
While the benefits are substantial, the integration of AI and FL into petroleum pro-
cessing is not without challenges. Issues such as data quality, algorithm bias, the
need for skilled personnel, and cybersecurity risks pose significant hurdles (Aldoseri
et al., 2023). Additionally, the high initial investment and the complexity of inte-
grating these technologies into existing systems can be daunting for some operators
(Rath et al., 2024). The future direction of AI and FL in the petroleum industry is
geared toward overcoming these challenges through continued innovation, collabo-
ration, and regulation. As these technologies mature and their applications expand,
the petroleum industry is poised to become more efficient, safer, and more environ-
mentally sustainable. The ongoing digital transformation, powered by AI and FL,
is set to redefine the industry, promising a future where petroleum processing not
only meets the global energy demand but does so more responsibly and sustainably
(Ahmad et al., 2022).
Previous literature discusses the basis and concepts of FL. Chiaro et al. (2023)
introduced FL-Enhance, an innovative federated learning approach leveraging
conditional generative adversarial networks (cGANs). This method ensures robust
privacy protection, resisting differential privacy (DP) and model inversion attacks,
all while maintaining competitive model performance relative to traditional FL
methods. Jin et al. (2024) introduced FL-IIDS, a new IDS framework designed to
136 Artificial Intelligence Using Federated Learning

effectively tackle the issue of catastrophic forgetting. Agiollo et al. (2024) debuted
EneA-FL as an innovative serverless federated learning framework, distinguish-
ing itself with a sophisticated energy management component designed specifically
for clients operating under resource limitations. Greguric et al. (2024) explored the
implementation of federated learning within a setting characterized by mixed traf-
fic flow, incorporating connected and automated vehicles (CAVs). Rafi et al. (2024)
introduced a detailed survey of privacy and fairness concerns in the context of FL.
De Rango et al. (2023) reported the architecture and theoretical equations of feder-
ated learning in detail.
Federated learning in the context of petroleum processes involves leveraging dis-
tributed data sources across various oil rigs, refineries, or other facilities to improve
machine learning models while preserving data privacy and security. These equa-
tions illustrate the basic mechanics of federated learning in the context of petroleum
processes, where models are trained across distributed data sources while preserving
data privacy and security. Here’s a simplified equation representing the federated
learning process:

1. Global Model Update Equation:

0t +1 = 0t − n.VL (0t ; Dt ) (1)

2. Local Model Update Equation:

0ti+1 = 0t − n.VL (0t ; Dti ) (2)

3. Aggregation Equation:
1
E
N
0t+1 = 0i (3)
N i =1 t +1

This chapter explores the application of AI and FL in petroleum processing,

highlighting their impact on optimizing operations, predictive maintenance, and
reducing carbon footprint. The flow chart and chapter organization are provided
in Figure 8.1.

AI IN PETROLEUM PROCESSING
AI technologies, including machine learning, deep learning, and computer vision,
have been increasingly applied in the petroleum sector to analyze data, automate
processes, and make predictive decisions. The integration of artificial intelligence
into petroleum processing signifies a revolutionary shift towards more efficient, sus-
tainable, and safer operations within the oil and gas industry (Hussain et al., 2024).
AI encompasses a range of technologies, including machine learning, deep learning,
natural language processing, and robotics, which can be leveraged to optimize vari-
ous aspects of the petroleum value chain. Here, we delve into the critical applica-
tions, benefits, and challenges of AI in petroleum processing, shedding light on its
transformative impact.
Application of AI and FL in Petroleum Processing 137

Critical Predictive
Applications of AI Decentralized Maintenance for
in Petroleum Data Utilization Oil Rigs and
Processing Equipment

Reservoir
Benefits of AI in Preserving Data
Management and
Petroleum Privacy and
Production
Processing Security
Optimization

Challenges of AI in Safety and

Real-Time
Petroleum Environmental
Adaptability
Processing Monitoring

Optimization of Energy Efficiency

Operational and Emissions
Processes Reduction

Improving
Resource
Refinery Safety
Efficiency
with FL

Collaborative
Innovation

Compliance with
Regulatory
Requirements

FIGURE 8.1 Flow chart of current research progress.

Critical Applications of AI in Petroleum Processing

The integration of artificial intelligence into petroleum processing marks a signifi-
cant evolution in how the industry approaches its most complex and critical chal-
lenges. AI’s capabilities extend across the entire petroleum value chain, offering
innovative solutions that enhance efficiency, safety, and sustainability. AI technolo-
gies significantly enhance the exploration and production phase by improving the
accuracy of seismic data interpretation. Machine learning algorithms can analyze
seismic and geological data to identify potential hydrocarbon reserves with greater
precision (Wang et al., 2023). This capability not only reduces the risk and cost
associated with drilling but also minimizes environmental impact by targeting drill-
ing operations more effectively. Furthermore, predictive maintenance powered by
AI algorithms can foresee equipment failures before they occur by analyzing his-
torical data, sensor readings, and real-time monitoring (Molęda et al., 2023). This
138 Artificial Intelligence Using Federated Learning

predictive approach ensures maintenance activities are only performed when neces-
sary, reducing downtime, extending equipment life, and saving costs associated with
unplanned maintenance and production losses. In addition, AI models can optimize
refining processes by continuously analyzing operational data and adjusting pro-
cessing parameters in real time. This optimization includes controlling the tempera-
ture, pressure, and chemical inputs to maximize yield, improve product quality, and
enhance energy efficiency. AI-driven process optimization leads to significant cost
savings and operational efficiencies. AI applications extend to improving safety and
environmental monitoring by employing sensors and computer vision to detect leaks,
hazardous conditions, or equipment failures (Pishgar et al., 2021). These technolo-
gies enable early detection and intervention, reducing the risk of accidents, minimiz-
ing environmental harm, and ensuring compliance with safety and environmental
regulations.

Benefits of AI in Petroleum Processing

AI optimizes operations, reduces energy consumption, and maximizes yield,
directly contributing to increased operational efficiency (Ahmad et al., 2021; Salem
et al., 2023). Furthermore, with AI, decision-makers can gain deeper insights into
their operations, leading to more informed and strategic decisions. By improving
predictive maintenance and process optimization, AI technologies can significantly
reduce operational and maintenance costs. AI aids in reducing the environmental
impact of petroleum processing by optimizing energy use and detecting leaks or
hazardous emissions early, thereby preventing pollution (Nemitallah et al., 2023).
In petroleum processing, vast amounts of data are generated from seismic surveys,
drilling operations, and daily processing activities. AI algorithms can process this
data to identify patterns, trends, and anomalies (El-hoshoudy et al., 2022; Himeur
et al., 2021). For instance, machine learning models can analyze seismic data to
predict the presence of oil reserves with higher accuracy, reducing the risk and cost
associated with exploration. Finally, AI-driven predictive maintenance models can
forecast equipment failures before they occur by analyzing historical operation data,
sensor readings, and real-time monitoring (Ayvaz & Alpay, 2021). This capability
allows for timely maintenance, reducing downtime and extending the lifespan of
critical infrastructure in refineries. AI algorithms optimize petroleum processing
operations by adjusting variables such as temperature, pressure, and flow rates in real
time based on feedstock characteristics and desired product specifications (Pomeroy
et al., 2022). This optimization enhances energy efficiency, reduces waste, and maxi-
mizes yield.

Challenges of AI in Petroleum Processing

The application of AI in petroleum processing offers remarkable opportunities to
enhance efficiency, safety, and sustainability. As the industry continues to embrace
digital transformation, AI technologies are set to play a pivotal role in revolution-
izing petroleum processing operations. Overcoming challenges related to data, inte-
gration, cybersecurity, and workforce adaptation will be crucial to fully realizing
Application of AI and FL in Petroleum Processing 139

the potential of AI in this sector (Campion et al., 2022). The continued advance-
ment and integration of AI technologies promise to drive significant improvements
in petroleum processing, setting the stage for a more efficient, safe, and sustainable
industry. The effectiveness of AI is heavily dependent on the quality and quantity of
data. Inconsistent, incomplete, or poor-quality data can hinder the performance of
AI models (Gomaa et al., 2022; Munappy et al., 2022). Integrating AI technologies
with existing infrastructure and systems poses technical and operational challenges,
requiring significant investment and expertise. Increased reliance on AI and digi-
tal technologies raises cybersecurity concerns, necessitating robust security mea-
sures to protect sensitive data and operational technology (Abdel-Rahman, 2023).
The adoption of AI requires a workforce skilled in data science, machine learning,
and AI technologies, alongside ongoing training and development programs (Gouda
et al., 2022; Jain et al., 2023).

FEDERATED LEARNING IN PETROLEUM PROCESSING

Federated learning is a machine learning approach that allows for model training
across multiple decentralized edge devices or servers holding local data samples,
without the need to exchange raw data. In FL, instead of aggregating data into a
central server for model training, the model is trained locally on each device using
its respective data. Only model updates (gradients) are sent to a central server, where
they are aggregated to update the global model (Qi et al., 2024). This approach
preserves data privacy and security since raw data never leaves the local devices.
Federated learning offers a promising approach to leverage AI in petroleum process-
ing while addressing data privacy and security concerns. FL is particularly useful in
scenarios where data is sensitive, such as healthcare, finance, or industries like petro-
leum processing, where companies may be reluctant to share proprietary data due to
privacy, security, or competitive reasons. FL allows for collaborative model training
while respecting data ownership and privacy constraints. It enables companies to
leverage the collective intelligence of distributed data sources without compromis-
ing data security or violating privacy regulations (Zhuang et al., 2023). FL has the
potential to drive advancements in various fields by enabling collaborative innova-
tion and knowledge sharing while addressing privacy concerns. Federated learning
holds considerable promise for revolutionizing various aspects of petroleum process-
ing, offering unique advantages and opportunities for the industry.
Chena et al. (2021) employed FL for enhancing the efficacy and adaptability of
the oil/water layer model. As depicted in Figure 8.2, during each round of FL train-
ing, symbolized by a block, individual clients train a model using their local data.
After filtering the model, it is then transmitted to a central server. Subsequently,
the central server constructs a global model through dynamic weighted fusion and
distributes it back to each client for the subsequent iteration. This process ensures
the elimination of data leakage from the block since only model transmission occurs
without data communication.
Federated learning presents a compelling solution for decentralized data utiliza-
tion in petroleum processing, overcoming challenges related to data silos, geographi-
cal dispersion, privacy, security, and operational agility. By enabling models to be
140 Artificial Intelligence Using Federated Learning

FIGURE 8.2 FL framework for oil/water layer identification based on dynamic weighted
fusion (Chena et al., 2021).

trained directly on distributed data sources while preserving data privacy and mini-
mizing latency, FL empowers the industry to unlock actionable insights, optimize
processes, and drive innovation collaboratively and securely.

Decentralized Data Utilization

Petroleum processing involves a plethora of data sources scattered across different
stages such as exploration, extraction, refining, and distribution. FL allows models to
be trained directly on this distributed data without needing to centralize it, overcom-
ing challenges associated with data silos and geographical dispersion (Khan et al.,
2021). Decentralized data utilization is a critical aspect of modernizing petroleum
processing, and Federated learning offers a potent solution to effectively harness the
wealth of distributed data sources inherent to the industry (Shah et al., 2022). Here’s
a discussion on how FL addresses the challenges associated with decentralized data
in petroleum processing. Petroleum processing operations generate vast amounts of
data across various stages, including exploration, extraction, refining, and distribu-
tion. Traditionally, this data tends to be siloed within different departments or loca-
tions, hindering comprehensive analysis and decision-making. FL enables models
to be trained directly on data distributed across these silos, breaking down barriers
Application of AI and FL in Petroleum Processing 141

and allowing for holistic insights that encompass the entire process chain. Petroleum
processing facilities are often geographically dispersed, with operations spanning
remote locations, offshore platforms, and refinery complexes. This geographical dis-
persion presents challenges in aggregating and analyzing data centrally. FL decen-
tralizes the training process, allowing models to be developed and updated directly
on-site or within the vicinity of data sources. This approach minimizes latency and
bandwidth constraints associated with transferring data to a centralized location for
analysis. The petroleum industry deals with highly sensitive and proprietary data,
including geological surveys, drilling logs, production metrics, and market intelli-
gence. Centralizing this data for analysis raises significant privacy and security con-
cerns, as it increases the risk of unauthorized access or breaches. FL addresses these
concerns by keeping data localized on devices or servers, ensuring that sensitive
information remains protected while still allowing for collaborative model training
and knowledge sharing. FL enables real-time model training and inference directly
on decentralized data sources, enhancing operational efficiency and agility in petro-
leum processing (Rane & Narvel, 2022). By leveraging FL, companies can develop
AI models that adapt to dynamic conditions, such as changing market demands, geo-
logical variations, or regulatory requirements. This capability empowers decision-
makers with timely insights and recommendations, enabling proactive responses to
emerging challenges and opportunities (Nguyen et al., 2022). FL offers scalability
and resilience advantages by distributing computational workload and training tasks
across a network of devices or servers. This distributed approach reduces the risk of
single points of failure and enhances the overall robustness of AI systems deployed
in petroleum processing. Additionally, FL facilitates the integration of new data
sources and sensors into existing models, allowing for continuous improvement and
adaptation over time (Beltrán et al., 2024).

Preserving Data Privacy and Security

FL addresses data privacy and security concerns by keeping sensitive information
localized on decentralized devices or servers. This approach ensures that proprietary
data, such as drilling logs, production rates, and reservoir characteristics, remains
protected while still enabling collaborative model training and improvement. Several
key principles and practices are essential for effectively safeguarding data privacy
and security. FL collects and retains only the minimum amount of data necessary to
fulfill a specific purpose. Minimizing data collection reduces the risk of unauthorized
access and misuse of sensitive information. On the other hand, FL encrypts data both
in transit and at rest to protect it from unauthorized access. Strong encryption algo-
rithms ensure that even if data is intercepted, it remains unintelligible to unauthorized
parties. Regarding access control, FL implements robust access control mechanisms
to restrict data access to authorized users only. Role-based access control (RBAC),
multi-factor authentication (MFA), and least privilege principles help ensure that
individuals only have access to the data necessary for their role or task (Omotunde &
Ahmed, 2023). Moreover, it anonymizes or pseudonymizes personally identifiable
information (PII) to prevent the identification of individuals. By removing or obfus-
cating identifying information, organizations can reduce the risk of privacy breaches
142 Artificial Intelligence Using Federated Learning

while still retaining valuable data for analysis and research purposes. Regarding
data governance, FL establishes clear policies and procedures for managing data
throughout its lifecycle. Data governance frameworks define roles and responsi-
bilities, data classification criteria, retention periods, and data disposal protocols
to ensure consistent and compliant data handling practices. FL allows compliance
with regulations through updating Stay informed about relevant data privacy regula-
tions and ensure compliance with applicable laws and standards. Regulations such as
the General Data Protection Regulation (GDPR), California Consumer Privacy Act
(CCPA), and Health Insurance Portability and Accountability Act (HIPAA) impose
strict requirements for data protection and privacy, and non-compliance can result
in severe penalties. By adhering to these principles and implementing appropriate
technical and organizational measures, organizations can effectively preserve data
privacy and security, build trust with stakeholders, and mitigate the risks associated
with unauthorized access, data breaches, and regulatory non-compliance.

Real-Time Adaptability
FL facilitates the training and updating of AI models in real time, enabling rapid
adaptation to dynamic conditions within the petroleum processing environment.
This capability is particularly valuable in responding to fluctuating market demands,
changing regulatory requirements, and evolving operational challenges. Real-time
adaptability is a crucial aspect of any modern system, particularly in industries
like petroleum processing where conditions can change rapidly and unpredictably.
Federated learning offers a promising solution to this challenge by enabling the
training and updating of AI models directly on the devices or servers where the
data is generated, rather than relying on centralized data repositories. In the context
of petroleum processing, FL allows AI models to continuously learn and adapt to
dynamic conditions within the environment (Nguyen et al., 2021). FL facilitates rapid
response to market demands by allowing AI models to quickly adapt to fluctuating
market demands. For example, if there is a sudden increase in the demand for a par-
ticular petroleum product, FL allows the model to incorporate new data and adjust
its predictions or recommendations accordingly, ensuring optimal production and
distribution strategies. In addition, FL allows compliance with regulatory require-
ments in the petroleum industry which can change frequently and may vary depend-
ing on factors such as location, environmental concerns, and safety standards. FL
enables AI models to stay up-to-date with these regulations by continuously incor-
porating new data and adjusting their decision-making processes to ensure compli-
ance. Finally, FL allows addressing operational challenges in petroleum processing
environments which are inherently complex, with numerous variables that can affect
production efficiency, product quality, and safety. FL allows AI models to adapt in
real time to evolving operational challenges by learning from new data and updating
their algorithms accordingly. For example, if there is a sudden equipment malfunc-
tion or a change in raw material composition, FL enables the model to adjust its
parameters to optimize performance and minimize disruptions. The ability of FL to
facilitate real-time adaptability in petroleum processing can lead to improved effi-
ciency, flexibility, and responsiveness in operations. By continuously learning from
Application of AI and FL in Petroleum Processing 143

new data and updating their models on the fly, organizations can stay competitive in
a rapidly changing industry landscape while also ensuring compliance with regula-
tions and delivering value to customers.

Optimization of Operational Processes

By leveraging FL, petroleum companies can develop AI models tailored to optimize
various operational processes, including reservoir management, well production
optimization, refinery efficiency, and supply chain logistics. These models can help
streamline workflows, enhance decision-making, and ultimately improve operational
performance and profitability. Optimization of operational processes is essential for
petroleum companies to maximize efficiency, reduce costs, and enhance overall per-
formance. Federated learning offers a powerful approach to achieving these goals by
enabling the development of tailored AI models that can optimize various aspects of
petroleum operations (Boobalan et al., 2022). Regarding reservoir management, FL
allows petroleum companies to develop AI models that analyze vast amounts of data
from multiple sources, including geological surveys, well logs, and production data,
to optimize reservoir management strategies. These models can predict reservoir
behavior, identify optimal drilling locations, and optimize production techniques to
maximize recovery rates while minimizing costs and environmental impact. Also,
FL enables the development of AI models that continuously analyze data from indi-
vidual wells, such as production rates, pressure levels, and equipment status, to opti-
mize production operations. These models can detect anomalies, predict equipment
failures, and recommend optimal operating parameters to maximize production effi-
ciency and minimize downtime. On refinery processes, FL can be used to develop
AI models that optimize refinery operations, such as crude oil blending, process
control, and energy management. By analyzing real-time data from sensors and con-
trol systems, these models can identify opportunities to improve energy efficiency,
reduce emissions, and optimize product yields, leading to lower operating costs and
increased profitability. Generally, by leveraging FL to develop tailored AI models
for various operational processes, petroleum companies can streamline workflows,
enhance decision-making, and ultimately improve operational performance and
profitability. These models enable companies to make data-driven decisions in real
time, adapt to changing market conditions, and optimize operations at every stage of
the petroleum value chain.

Resource Efficiency
FL minimizes the need for extensive data transfer and centralization, resulting
in more efficient utilization of network bandwidth and computational resources.
This efficiency is especially beneficial in remote or offshore locations where con-
nectivity may be limited, allowing for on-device model training and inference
without reliance on constant internet connectivity. Resource efficiency is a critical
factor in any technological solution, especially in industries like petroleum pro-
cessing where operations often occur in remote or offshore locations with limited
connectivity. Federated learning presents a solution that minimizes the need for
144 Artificial Intelligence Using Federated Learning

extensive data transfer and centralization, thereby optimizing the utilization of

network bandwidth and computational resources (Pinder, 2001). FL operates by
training machine learning models locally on devices or servers where the data is
generated, rather than transferring large volumes of raw data to a central location.
This approach significantly reduces the need for data transfer over networks, par-
ticularly in remote or offshore locations where bandwidth may be limited or expen-
sive. By keeping data local, FL minimizes latency and bandwidth usage, allowing
for more efficient utilization of network resources. Furthermore, FL enables on-
device model training and inference, meaning that AI models can be trained and
executed directly on devices or servers at the edge of the network, without relying
on constant internet connectivity or access to a centralized server. This capabil-
ity is especially beneficial in remote or offshore locations where internet connec-
tivity may be intermittent or unreliable. By performing computations locally, FL
reduces dependence on centralized infrastructure and enhances the efficiency of
resource utilization. In addition, FL inherently prioritizes data privacy and secu-
rity by keeping sensitive data local and only transmitting model updates or aggre-
gated information to a central server. This decentralized approach minimizes the
risk of data breaches or unauthorized access to sensitive information, particularly
in environments where data protection regulations are stringent. By ensuring
data privacy and security, FL enables petroleum companies to leverage valuable
data assets without compromising confidentiality or regulatory compliance. FL is
highly scalable and flexible, allowing organizations to deploy AI models across
distributed networks of devices or servers with minimal overhead. This scalability
enables petroleum companies to adapt to changing operational requirements and
scale their AI initiatives as needed, without significant investments in centralized
infrastructure. By distributing computational tasks across edge devices, FL maxi-
mizes resource utilization and ensures optimal performance even in challenging
environments. Generally, FL enhances resource efficiency in petroleum process-
ing by minimizing the need for extensive data transfer and centralization, enabling
on-device model training and inference, prioritizing data privacy and security, and
offering scalability and flexibility to adapt to diverse operational requirements. By
leveraging FL, petroleum companies can optimize the utilization of network band-
width and computational resources, even in remote or offshore locations, thereby
enhancing operational efficiency and reducing costs.

Collaborative Innovation
FL fosters collaboration among different stakeholders in the petroleum industry,
including oil companies, service providers, research institutions, and regulatory
agencies. By enabling collective learning from decentralized data sources while
respecting data ownership and privacy constraints, FL accelerates innovation and
knowledge sharing across the industry ecosystem. Collaborative innovation repre-
sents a strategic approach that leverages the collective knowledge, resources, and
capabilities of various stakeholders to drive innovation and create value in ways that
would be challenging for any single entity to achieve independently. This concept is
crucial across industries and sectors, including technology, healthcare, education, and
Application of AI and FL in Petroleum Processing 145

environmental sustainability, among others (Capobianco et al., 2021). Let’s explore

the key aspects of collaborative innovation, its benefits, challenges, and implications
for future developments. Collaborative innovation involves multiple stakeholders,
such as businesses, research institutions, government agencies, and sometimes even
competitors, working together. The diversity of perspectives, expertise, and resources
contributes to more holistic and innovative solutions. Many collaborative innovation
initiatives rely on open platforms that allow external contributors to offer ideas, solu-
tions, and feedback. These platforms can significantly expand the innovation poten-
tial beyond the internal capacities of organizations. At the heart of collaborative
innovation is the principle of co-creation, where stakeholders actively participate in
the creation process, leading to shared value. This approach fosters a deeper sense
of ownership and commitment to the innovation outcomes. Collaborative innova-
tion often involves partnerships across different sectors, bringing together public and
private entities, academia, and non-profits to address complex challenges that span
beyond the scope of any single sector. Collaborative innovation offers numerous ben-
efits that can significantly enhance the capacity for problem-solving, creativity, and
value creation across industries and sectors. By bringing together diverse stakehold-
ers, such as companies, research institutions, governments, and communities, collab-
orative innovation fosters a synergistic environment where the sum of contributions
exceeds what any single entity could achieve alone. By pooling resources and knowl-
edge, collaborative innovation can significantly speed up the development process,
leading to quicker deployment of new technologies and solutions. The convergence
of diverse expertise and viewpoints fosters creativity, leading to more effective and
comprehensive problem-solving strategies. Collaboration allows partners to share
the risks associated with innovation, making it easier to undertake ambitious proj-
ects and explore new territories. Partnerships can open up new markets and access
to cutting-edge technologies, providing a competitive edge and fostering growth.
Collaborative innovation, while offering numerous benefits, also comes with its set
of challenges. These challenges can impact the efficiency, effectiveness, and over-
all success of collaborative efforts (Zelenika & Pearce, 2014). Understanding and
addressing these challenges is crucial for organizations and individuals aiming to
harness the full potential of collaborative innovation. As the world becomes increas-
ingly interconnected and the pace of technological change accelerates, collaborative
innovation is likely to play an even more critical role in addressing global chal-
lenges and driving progress. Collaborative efforts are increasingly focusing on sus-
tainable development goals, recognizing that addressing environmental and social
challenges requires joint action. The development and adoption of advanced digi-
tal collaboration platforms will facilitate more seamless and efficient cooperation
among stakeholders. There’s a growing recognition of the importance of inclusivity
in innovation processes, ensuring that diverse voices and perspectives are heard and
valued. Collaborative innovation offers a promising pathway to address the com-
plex challenges of today’s world, leveraging the strengths and capabilities of diverse
stakeholders. While it comes with its set of challenges, the potential benefits in terms
of accelerated innovation, risk sharing, and access to new opportunities make it an
increasingly attractive strategy. As we move forward, fostering environments that
support collaboration, addressing IP and management challenges, and leveraging
146 Artificial Intelligence Using Federated Learning

technology for better collaboration will be key to realizing the full potential of col-
laborative innovation (Zhang et al., 2019).

Compliance with Regulatory Requirements

FL offers a framework for AI deployment that aligns with stringent regulatory man-
dates governing the petroleum sector. By minimizing data exposure and ensuring
compliance with privacy regulations such as GDPR and industry-specific standards
like those set forth by the American Petroleum Institute (API), FL helps mitigate
regulatory risks and enhances overall governance and transparency (Asad, 2023).
Compliance with regulatory requirements is a critical aspect of collaborative innova-
tion, particularly in industries that are heavily regulated, such as healthcare, finance,
pharmaceuticals, and energy. The complexity of regulatory landscapes across dif-
ferent regions and sectors can pose significant challenges to collaborative projects.
However, adherence to these regulations is essential not only for legal compliance but
also for maintaining public trust and protecting consumer rights. Collaborative proj-
ects often span multiple jurisdictions, each with its own set of regulations. A com-
prehensive understanding of the regulatory requirements in all relevant regions is
crucial to avoid legal pitfalls. Different industries are subject to specific regulatory
frameworks. Collaborators must be aware of and adhere to the regulations that apply
to their sector, whether it involves data protection, clinical trials, financial reporting,
or environmental standards.
Ensuring compliance with regulatory requirements, particularly in collaborative
innovation projects that may span multiple industries and jurisdictions, is a com-
plex but critical task. Effective strategies for compliance help manage risks, protect
intellectual property, maintain data privacy, and ensure product safety, among other
regulatory concerns. Establishment of a regulatory compliance team responsible
for understanding, monitoring, and ensuring compliance with all relevant regula-
tions. This team should include legal experts and regulatory specialists familiar with
the specific requirements of the industries and regions involved. Moreover, regular
audits can help identify potential compliance issues before they become problematic.
These should assess all aspects of the project, from data management to product
safety, to ensure that every element meets regulatory standards. In the same regard, a
comprehensive compliance plan should outline how the collaboration will adhere to
regulatory requirements. This includes procedures for data handling, privacy protec-
tion, financial reporting, and any other relevant areas. The plan should be regularly
reviewed and updated as regulations change or the project evolves. Many regula-
tions, especially those concerning privacy (such as GDPR in Europe or HIPAA in
the United States), focus on data protection. Implementing secure data management
practices, ensuring data is encrypted, and managing access rights meticulously
are crucial steps in compliance. Proactively engaging with regulatory bodies can
provide valuable insights into compliance requirements and help avoid misunder-
standings or conflicts. This engagement can include seeking advice, submitting
project plans for review, or participating in regulatory sandbox programs, where
available. Finally, training and awareness programs ensure that all participants in
the collaborative project are aware of the regulatory requirements and understand
Application of AI and FL in Petroleum Processing 147

their roles in maintaining compliance. Regular training sessions can keep everyone
updated on the latest regulations and best practices. Regulatory compliance is essen-
tial for businesses and organizations operating in various industries. Compliance
with regulations ensures that companies adhere to legal requirements set forth by
governments and regulatory bodies. While compliance efforts can sometimes be
seen as burdensome, they offer numerous benefits that ultimately contribute to the
success and sustainability of businesses (Griffith, 2015). Compliance with regula-
tory requirements is a foundational element of successful collaborative innovation.
It ensures that innovations not only meet legal standards but are also safe, reliable,
and trustworthy. While navigating the regulatory landscape can be complex, a pro-
active, informed approach to compliance can facilitate smoother project execution
and contribute to the long-term success and sustainability of collaborative innova-
tions. Benefits of regulatory compliance include: (1) Risk Mitigation: Compliance
reduces the risk of legal penalties, financial losses, and reputational damage that
can result from regulatory violations. (2) Market Access: Meeting regulatory stan-
dards is often a prerequisite for accessing certain markets or dealing with particular
customer segments. (3) Competitive Advantage: Demonstrating a commitment to
compliance can be a competitive differentiator, building trust with customers, part-
ners, and regulators. Federated learning emerges as a transformative approach to
harnessing artificial intelligence in petroleum processing, offering a decentralized,
privacy-preserving framework for data-driven decision-making, operational opti-
mization, and collaborative innovation. By leveraging the collective intelligence of
distributed data sources, FL holds immense promise for driving significant advance-
ments across various dimensions within the petroleum industry. FL’s decentralized
nature enables companies to collaborate effectively while respecting data privacy
and ownership, thereby overcoming traditional barriers to data sharing and collabo-
ration. This collaborative approach fosters a more open environment for innovation,
where stakeholders across the industry ecosystem can contribute their expertise and
insights without compromising sensitive information. Moreover, FL addresses criti-
cal concerns related to privacy, security, and regulatory compliance, ensuring that
data remains protected throughout the learning process. By adhering to strict pri-
vacy-preserving protocols and implementing robust security measures, FL enables
companies to derive actionable insights from their data while mitigating the risk
of unauthorized access or breaches. The potential benefits of FL in the petroleum
industry are vast. From optimizing production processes and predicting equipment
failures to improving environmental sustainability and reducing operational costs,
FL can drive efficiency gains and foster sustainable practices across the industry
(Bibri et al., 2024). Furthermore, FL has the potential to enhance the competitiveness
of companies operating in the petroleum sector. By enabling faster innovation cycles,
more informed decision-making, and the development of cutting-edge technologies,
FL empowers companies to stay ahead of the curve in a rapidly evolving market
landscape. In essence, Federated Learning represents a paradigm shift in how AI is
leveraged within the petroleum industry, offering a pathway to unlock new oppor-
tunities for efficiency, sustainability, and competitiveness. By embracing FL and its
collaborative ethos, companies can position themselves at the forefront of innova-
tion, driving positive change and unlocking value across the industry ecosystem.
148 Artificial Intelligence Using Federated Learning

CASE STUDIES OF FEDERATED LEARNING

IN PETROLEUM PROCESSING
Specific case studies of federated learning in petroleum processing are relatively
limited due to the emerging nature of FL technology and the sensitivity of data
within the petroleum industry. However, we can discuss hypothetical scenarios and
potential applications where FL could be beneficial in petroleum processing (Pham
et al., 2021). In the following discussion, short hints on the supposed case studies of
FL in petroleum processing are discussed.

Predictive Maintenance for Oil Rigs and Equipment

In this scenario, FL could be applied to predict equipment failures and maintenance
needs across multiple oil rigs operated by different companies. Each rig could locally
train predictive maintenance models using sensor data, maintenance logs, and his-
torical failure data. By aggregating model updates from various rigs without sharing
raw data, a global predictive maintenance model could be trained. This model would
provide insights into potential equipment failures and maintenance schedules while
preserving the privacy of sensitive operational data (Mobley, 2002).

Reservoir Management and Production Optimization

FL could facilitate collaboration among multiple oil companies to optimize reser-
voir management and production strategies while protecting proprietary geologi-
cal and production data. Each company could train local models using reservoir
data, well performance metrics, and production history. These models could then be
aggregated to develop a global model for reservoir management, helping optimize
production rates, well placement, and injection strategies (Huseby & Haavardsson,
2009). A leading petroleum company implemented AI-driven seismic data analysis
to identify potential oil reserves. The application of deep learning models increased
the accuracy of predictions by 30%, significantly reducing exploration costs and
environmental impact. While these examples are hypothetical, they illustrate the
potential for FL to drive innovation and collaboration in petroleum processing while
addressing privacy, security, and regulatory concerns. As FL technology contin-
ues to mature and industry stakeholders become more receptive to collaborative
approaches, we may see real-world case studies and implementations of FL in petro-
leum processing emerge in the future (Konečný et al., 2016).

Safety and Environmental Monitoring

FL could enable collaborative monitoring of safety and environmental parameters
across oil refineries and processing facilities operated by different companies. Local
models could be trained using sensor data to detect anomalies, leaks, or environ-
mental hazards within each facility. Aggregating model updates would allow for the
development of a global monitoring system, enhancing safety and environmental
compliance across the industry (Siddique et al., 2024).
Application of AI and FL in Petroleum Processing 149

Energy Efficiency and Emissions Reduction

FL could support efforts to improve energy efficiency and reduce emissions in
petroleum processing operations. Companies could locally train models to optimize
energy usage, minimize flaring, and reduce greenhouse gas emissions. By sharing
model updates through FL, best practices and insights could be disseminated across
the industry, leading to collective improvements in energy efficiency and environ-
mental performance (Qiu et al., 2023).

Improving Refinery Safety with FL

A consortium of petroleum processing companies adopted an FL approach to train
anomaly detection models across their refineries. This initiative allowed for real-
time monitoring and early detection of potential hazards, decreasing the incidence
of accidents by 20%. Improving refinery safety is a paramount concern within the
petroleum industry, given the potentially hazardous nature of refinery operations and
the importance of ensuring the well-being of workers, communities, and the envi-
ronment. Federated learning presents a promising approach to enhancing refinery
safety by leveraging decentralized data sources while respecting privacy and secu-
rity considerations. FL enables collaborative model training using data from vari-
ous refinery components, such as sensors, equipment logs, and maintenance records,
without sharing raw data. Local models can be trained on each refinery’s data to
detect anomalies and predict equipment failures or maintenance needs. By aggregat-
ing model updates through FL, a global anomaly detection and predictive mainte-
nance model can be developed, providing early warnings for potential safety hazards
and allowing for proactive maintenance to prevent accidents (de Castro Ferreira,
2023). Moreover, FL facilitates collaborative monitoring of environmental param-
eters, such as air quality, emissions, and effluent discharges, across multiple refiner-
ies. Local models trained on refinery-specific environmental data can help identify
emission hotspots, optimize processes to reduce pollution, and ensure compliance
with regulatory standards. By sharing model updates through FL, refineries can col-
lectively improve environmental performance and mitigate risks to public health
and the environment (Junaid et al., 2023). In addition, FL can support the develop-
ment of personalized safety training programs for refinery workers based on their
roles, experience levels, and safety records. Local models trained on workers’ safety
incident data can identify patterns and trends to tailor training materials and inter-
ventions effectively. Additionally, FL can facilitate real-time incident response by
enabling the aggregation of safety incident data from multiple refineries to identify
common root causes and share best practices for prevention and response. FL allows
refineries to collaboratively optimize processes and manage safety risks by sharing
insights gleaned from operational data while preserving data privacy. Local models
trained on process data can identify operational inefficiencies, safety hazards, and
potential process deviations that may lead to safety incidents. By aggregating model
updates through FL, refineries can collectively identify and address safety risks,
enhance process safety, and improve overall operational performance. Finally, FL
supports collaborative efforts to ensure regulatory compliance by sharing insights
150 Artificial Intelligence Using Federated Learning

and best practices for meeting safety and environmental regulations. Local mod-
els trained on regulatory compliance data can help refineries identify areas of non-
compliance, implement corrective actions, and streamline reporting processes. By
leveraging FL, refineries can collaborate on compliance initiatives while protecting
sensitive regulatory data and maintaining transparency with regulatory authorities.
Federated learning offers a powerful framework for improving refinery safety by
enabling collaborative model training, data sharing, and insights generation while
addressing privacy, security, and regulatory concerns. By harnessing the collective
intelligence of distributed data sources, refineries can enhance safety practices, miti-
gate risks, and foster a culture of continuous improvement to ensure the safety and
well-being of workers, communities, and the environment.

CHALLENGES AND FUTURE DIRECTIONS

While AI and FL offer substantial benefits to the petroleum industry, they also
present challenges, including data quality, model interpretability, and the need for
skilled personnel. Future advancements in AI and FL technologies, coupled with
industry-specific solutions, are expected to address these challenges, further enhanc-
ing the efficiency, safety, and sustainability of petroleum processing operations.
Implementing AI and federated learning in the petroleum industry comes with its set
of challenges, ranging from data complexities to regulatory constraints. The petro-
leum industry deals with vast amounts of complex and heterogeneous data, includ-
ing geological surveys, drilling logs, production records, and sensor data. Ensuring
the quality, consistency, and availability of data remains a challenge, particularly in
remote or offshore locations. Despite the advancements in AI, the interpretability of
models remains a challenge. Understanding the decisions made by AI models, espe-
cially in critical areas such as reservoir management or safety monitoring, is essential
for gaining trust and ensuring accountability. Implementing AI and FL technologies
requires specialized expertise in data science, machine learning, and domain knowl-
edge of the petroleum industry. However, there is a shortage of skilled personnel
with expertise in both domains, hindering the adoption and implementation of these
technologies. The petroleum industry operates within a highly regulated environ-
ment, with stringent safety, environmental, and data privacy regulations. Ensuring
compliance with these regulations while leveraging AI and FL technologies poses
challenges related to data privacy, security, and regulatory reporting. Overcoming
these challenges requires concerted efforts from industry stakeholders, policymak-
ers, and technology providers. Addressing data quality issues, ensuring privacy and
security, achieving interpretability in AI models, navigating regulatory require-
ments, bridging the talent gap, and scaling infrastructure are key focus areas for
advancing AI and FL in the petroleum industry. Despite the challenges, the potential
benefits of leveraging AI and FL, including improved efficiency, safety, and sustain-
ability, make overcoming these obstacles worthwhile.
Several future directions and solutions can further enhance the application of AI
and Federated learning in the petroleum industry, addressing existing challenges
and unlocking new opportunities. Future advancements in data integration and man-
agement solutions, such as data lakes, digital twins, and advanced data analytics
Application of AI and FL in Petroleum Processing 151

platforms, will help address challenges related to data quality, consistency, and avail-
ability in the petroleum industry. Research efforts focusing on explainable AI (XAI)
techniques will enable the development of models that provide transparent and inter-
pretable insights. Techniques such as model-agnostic interpretability methods and
attention mechanisms will enhance the explainability of AI models. Investment in
education and training programs tailored to the needs of the petroleum industry
will help bridge the talent gap. Collaborative efforts between academia, industry,
and government institutions can facilitate the development of skilled personnel with
expertise in both data science and petroleum engineering. Collaboration between
industry stakeholders and regulatory bodies is essential for developing regulatory
frameworks and standards that accommodate the use of AI and FL technologies in
the petroleum industry. Establishing guidelines for data privacy, security, and ethi-
cal AI will promote responsible deployment and adoption of these technologies. The
development of industry-specific AI solutions and platforms tailored to the unique
challenges and requirements of the petroleum industry will drive innovation and
adoption. Collaborative initiatives among industry players, startups, and technology
providers will facilitate the development and deployment of these solutions.

CONCLUSION
The integration of artificial intelligence and federated learning marks a signifi-
cant milestone in the evolution of the petroleum industry, offering transforma-
tive solutions to longstanding challenges in efficiency, safety, and environmental
sustainability. This chapter has explored the potential applications of FL within
petroleum processing, highlighting its role in optimizing operations, predictive
maintenance, and reducing carbon footprint. By leveraging AI-driven analytics
and FL’s collaborative approach, the industry can achieve higher productivity,
safety standards, and environmental stewardship. AI’s capabilities extend across
various facets of petroleum processing, from exploration and production opti-
mization to predictive maintenance and environmental monitoring. Through
sophisticated algorithms and data-driven insights, AI enables refineries to oper-
ate more efficiently, minimize downtime, and reduce emissions. However, the full
potential of AI is unlocked when combined with FL, which enables collaborative
model training without compromising sensitive data. FL’s decentralized learn-
ing mechanism offers several advantages, including privacy-preserving model
training, real-time adaptability, and resource efficiency. By harnessing the col-
lective intelligence of distributed data sources, FL facilitates the seamless inte-
gration of insights from various processing plants, leading to more accurate and
globally applicable models. Case studies have demonstrated FL’s efficacy in pre-
dictive maintenance, reservoir management, safety monitoring, and regulatory
compliance. Despite the promises of AI and FL, challenges persist, including
data quality, interpretability, talent gap, and regulatory compliance. However,
these challenges present opportunities for future innovation and collaboration.
Solutions such as advanced data integration, explainable AI, education, regula-
tory frameworks, and industry-specific AI platforms can address these challenges
and drive the widespread adoption of AI and FL in petroleum processing. Looking
152 Artificial Intelligence Using Federated Learning

ahead, the future of AI and FL in the petroleum industry is bright, with continued
advancements, collaboration, and regulation paving the way for a more efficient,
safer, and sustainable industry. The ongoing digital transformation, fueled by AI
and FL, promises to redefine petroleum processing, ensuring that it meets global
energy demands while minimizing its environmental footprint. By embracing AI
and FL technologies, the petroleum industry is poised to usher in a new era of
innovation and progress, shaping a more resilient and responsible energy land-
scape for generations to come.

REFERENCES
Abdel-Rahman, M. (2023). Advanced cybersecurity measures in IT service operations and
their crucial role in safeguarding enterprise data in a connected world. Eigenpub
Review of Science and Technology, 7(1), 138–158.
Agiollo, A., Bellavista, P., Mendula, M., & Omicini, A. (2024). EneA-FL: Energy-aware orchestra-
tion for serverless federated learning. Future Generation Computer Systems, 154, 219–234.
Ahmad, T., Zhang, D., Huang, C., Zhang, H., Dai, N., Song, Y., & Chen, H. (2021). Artificial
intelligence in sustainable energy industry: Status Quo, challenges and opportunities.
Journal of Cleaner Production, 289, 125834.
Ahmad, T., Zhu, H., Zhang, D., Tariq, R., Bassam, A., Ullah, F., AlGhamdi, A. S., &
Alshamrani, S. S. (2022). Energetics systems and artificial intelligence: Applications of
industry 4.0. Energy Reports, 8, 334–361.
Aldoseri, A., Al-Khalifa, K. N., & Hamouda, A. M. (2023). Re-thinking data strategy and
integration for artificial intelligence: Concepts, opportunities, and challenges. Applied
Sciences, 13(12), 7082.
Asad, M. (2023). “Security Management in Cloud Computing for healthcare Data,” Dissertation,
pp. 77. DiVA, id: diva2:1810709 https://www.diva-portal.org/smash/record.jsf?pid=diva2
%3A1810709&dswid=-7919
Ayvaz, S., & Alpay, K. (2021). Predictive maintenance system for production lines in manu-
facturing: A machine learning approach using IoT data in real time. Expert Systems
with Applications, 173(6), 114598. doi: 10.1016/j.eswa.2021.114598.
Beltrán, E. T. M., Gómez, Á. L. P., Feng, C., Sánchez, P. M. S., Bernal, S. L., Bovet, G., Pérez,
M. G., Pérez, G. M., & Celdrán, A. H. (2024). Fedstellar: A platform for decentralized
federated learning. Expert Systems with Applications, 242, 122861.
Bibri, S. E., Krogstie, J., Kaboli, A., & Alahi, A. (2024). Smarter eco-cities and their leading-
edge artificial intelligence of things solutions for environmental sustainability: A com-
prehensive systematic review. Environmental Science and Ecotechnology, 19, 100330.
Boobalan, P., Ramu, S. P., Pham, Q.-V., Dev, K., Pandya, S., Maddikunta, P. K. R., Gadekallu,
T. R., & Huynh-The, T. (2022). Fusion of federated learning and industrial Internet of
Things: A survey. Computer Networks, 212, 109048.
Campion, A., Gasco-Hernandez, M., Jankin Mikhaylov, S., & Esteve, M. (2022). Overcoming
the challenges of collaboratively adopting artificial intelligence in the public sector.
Social Science Computer Review, 40(2), 462–477.
Capobianco, N., Basile, V., Loia, F., & Vona, R. (2021). Toward a sustainable decommission-
ing of offshore platforms in the oil and gas industry: A PESTLE analysis. Sustainability,
13(11), 6266.
Chena, B., Zenga, X., & Zhang, W. (2021). Federated learning for cross-block oil-water layer
identification. arXiv preprint arXiv:2112.14359.
Chiaro, D., Prezioso, E., Ianni, M., & Giampaolo, F. (2023). FL-Enhance: A federated learn-
ing framework for balancing non-IID data with augmented and shared compressed
samples. Information Fusion, 98, 101836.
Application of AI and FL in Petroleum Processing 153

de Castro Ferreira, J. P. (2023). A Federated Learning Platform for High Speed Distributed
Data Streams. FAUP Organization. https://sigarra.up.pt/faup/en/web_base.gera_
pagina?p_pagina=1004586.
De Rango, F., Guerrieri, A., Raimondo, P., & Spezzano, G. (2023). HED-FL: A hierarchical,
energy efficient, and dynamic approach for edge Federated Learning. Pervasive and
Mobile Computing, 92, 101804.
El-hoshoudy, A., Ahmed, A., Gomaa, S., & Abdelhady, A. (2022). An artificial neural net-
work model for predicting the hydrate formation temperature. Arabian Journal for
Science and Engineering, 1–10.
Gomaa, S., Soliman, A. A., Nasr, K., Emara, R., El-Hoshoudy, A., & Attia, A. M. (2022).
Development of artificial neural network models to calculate the areal sweep efficiency
for direct line, staggered line drive, five-spot, and nine-spot injection patterns. Fuel,
317, 123564.
Gouda, A., Gomaa, S., Attia, A., Emara, R., Desouky, S., & El-hoshoudy, A. (2022).
Development of an artificial neural network model for predicting the dew point pres-
sure of retrograde gas condensate. Journal of Petroleum Science and Engineering, 208,
109284.
Gregurić, M., Vrbanić, F., & Ivanjko, E. (2024). Impact of federated deep learning on vehi-
cle-based speed control in mixed traffic flows. Journal of Parallel and Distributed
Computing, 186, 104812.
Griffith, S. J. (2015). Corporate governance in an era of compliance. William & Mary Law
Review, 57, 2075.
Himeur, Y., Ghanem, K., Alsalemi, A., Bensaali, F., & Amira, A. (2021). Artificial intelli-
gence based anomaly detection of energy consumption in buildings: A review, current
trends and new perspectives. Applied Energy, 287, 116601.
Huseby, A. B., & Haavardsson, N. F. (2009). Multi-reservoir production optimization.
European Journal of Operational Research, 199(1), 236–251.
Hussain, M., Alamri, A., Zhang, T., & Jamil, I. (2024). Application of artificial intelligence
in the oil and gas industry. In Engineering Applications of Artificial Intelligence (pp.
341–373). Springer.
Jain, P., Tripathi, V., Malladi, R., & Khang, A. (2023). Data-driven artificial intelligence (AI)
models in the workforce development planning. In Designing Workforce Management
Systems for Industry 4.0 (pp. 159–176). CRC Press.
Jin, Z., Zhou, J., Li, B., Wu, X., & Duan, C. (2024). FL-IIDS: A novel federated learning-
based incremental intrusion detection system. Future Generation Computer Systems,
151, 57–70.
Junaid, M., Zhang, Q., Cao, M., & Luqman, A. (2023). Nexus between technology enabled
supply chain dynamic capabilities, integration, resilience, and sustainable performance:
An empirical examination of healthcare organizations. Technological Forecasting and
Social Change, 196, 122828.
Khaldi, M. K., Al-Dhaifallah, M., & Taha, O. (2023). Artificial intelligence perspectives:
A systematic literature review on modeling, control, and optimization of fluid catalytic
cracking. Alexandria Engineering Journal, 80, 294–314.
Khan, L. U., Saad, W., Han, Z., Hossain, E., & Hong, C. S. (2021). Federated learning for inter-
net of things: Recent advances, taxonomy, and open challenges. IEEE Communications
Surveys & Tutorials, 23(3), 1759–1799.
Konečný, J., McMahan, H. B., Yu, F. X., Richtárik, P., Suresh, A. T., & Bacon, D. (2016).
Federated learning: Strategies for improving communication efficiency. arXiv preprint
arXiv:1610.05492.
Koroteev, D., & Tekic, Z. (2021). Artificial intelligence in oil and gas upstream: Trends, chal-
lenges, and scenarios for the future. Energy and AI, 3, 100041.
154 Artificial Intelligence Using Federated Learning

Mobley, R. K. (2002). An Introduction to Predictive Maintenance. Elsevier.

Molęda, M., Małysiak-Mrozek, B., Ding, W., Sunderam, V., & Mrozek, D. (2023). From cor-
rective to predictive maintenance—A review of maintenance approaches for the power
industry. Sensors, 23(13), 5970.
Munappy, A. R., Bosch, J., Olsson, H. H., Arpteg, A., & Brinne, B. (2022). Data manage-
ment for production quality deep learning models: Challenges and solutions. Journal
of Systems and Software, 191, 111359.
Nemitallah, M. A., Nabhan, M. A., Alowaifeer, M., Haeruman, A., Alzahrani, F., Habib, M.
A., Elshafei, M., Abouheaf, M. I., Aliyu, M., & Alfarraj, M. (2023). Artificial intel-
ligence for control and optimization of boilers’ performance and emissions: A review.
Journal of Cleaner Production, 138109.
Nguyen, D. C., Ding, M., Pham, Q.-V., Pathirana, P. N., Le, L. B., Seneviratne, A., Li, J., Niyato,
D., & Poor, H. V. (2021). Federated learning meets blockchain in edge computing:
Opportunities and challenges. IEEE Internet of Things Journal, 8(16), 12806–12825.
Nguyen, Q.-T., Tran, T. N., Heuchenne, C., & Tran, K. P. (2022). Decision support systems
for anomaly detection with the applications in smart manufacturing: A survey and
perspective. In Machine Learning and Probabilistic Graphical Models for Decision
Support Systems (pp. 34–61). CRC Press.
Omotunde, H., & Ahmed, M. (2023). A comprehensive review of security measures in data-
base systems: Assessing authentication, access control, and beyond. Mesopotamian
Journal of CyberSecurity, 2023, 115–133.
Pham, Q.-V., Dev, K., Maddikunta, P. K. R., Gadekallu, T. R., & Huynh-The, T. (2021).
Fusion of federated learning and industrial Internet of Things: A survey. arXiv preprint
arXiv:2101.00798.
Pinder, D. (2001). Offshore oil and gas: Global resource knowledge and technological change.
Ocean & Coastal Management, 44(9–10), 579–600.
Pishgar, M., Issa, S. F., Sietsema, M., Pratap, P., & Darabi, H. (2021). REDECA: A novel
framework to review artificial intelligence and its applications in occupational safety
and health. International Journal of Environmental Research and Public Health,
18(13), 6705.
Pomeroy, B., Grilc, M., & Likozar, B. (2022). Artificial neural networks for bio-based chemi-
cal production or biorefining: A review. Renewable and Sustainable Energy Reviews,
153, 111748.
Qi, P., Chiaro, D., Guzzo, A., Ianni, M., Fortino, G., & Piccialli, F. (2024, January). Model
aggregation techniques in federated learning: A comprehensive survey. Future
Generation Computer Systems, 150, 272–293.
Qiu, X., Parcollet, T., Fernandez-Marques, J., Gusmao, P. P., Gao, Y., Beutel, D. J., Topal, T.,
Mathur, A., & Lane, N. D. (2023). A first look into the carbon footprint of federated
learning. Journal of Machine Learning Research, 24(129), 1–23.
Rafi, T. H., Noor, F. A., Hussain, T., & Chae, D.-K. (2024). Fairness and privacy preserving
in federated learning: A survey. Information Fusion, 105, 102198.
Rane, S. B., & Narvel, Y. A. M. (2022). Data-driven decision making with Blockchain-IoT
integrated architecture: A project resource management agility perspective of industry
4.0. International Journal of System Assurance Engineering and Management, 1–19.
Rath, K. C., Khang, A., & Roy, D. (2024). The role of Internet of Things (IoT) technology in
industry 4.0 economy. In Advanced IoT Technologies and Applications in the Industry
4.0 Digital Economy (pp. 1–28). CRC Press.
Rauniyar, A., Hagos, D. H., Jha, D., Håkegård, J. E., Bagci, U., Rawat, D. B., & Vlassov, V.
(2024, March 1). Federated learning for medical applications: A taxonomy, current
trends, challenges, and future research directions. IEEE Internet of Things Journal,
11(5), 7374–7398. doi: 10.1109/JIOT.2023.3329061.
Application of AI and FL in Petroleum Processing 155

Salem, K. G., Tantawy, M. A., Gawish, A. A., Gomaa, S., & El-hoshoudy, A. (2023).
Nanoparticles assisted polymer flooding: Comprehensive assessment and empirical
correlation. Geoenergy Science and Engineering, 226, 211753.
Shah, M., Kshirsagar, A., & Panchal, J. (2022). Applications of Artificial Intelligence (AI)
and Machine Learning (ML) in the Petroleum Industry. CRC Press.
Siddique, A. A., Alasbali, N., Driss, M., Boulila, W., Alshehri, M. S., & Ahmad, J. (2024).
Sustainable collaboration: Federated learning for environmentally conscious forest fire
classification in Green Internet of Things (IoT). Internet of Things, 25, 101013.
Singh, B. (2023). Federated learning for envision future trajectory smart transport system
for climate preservation and smart green planet: Insights into global governance and
SDG-9 (Industry, Innovation and Infrastructure). National Journal of Environmental
Law, 6(2), 6–17.
Wang, Q., Chen, D., Li, M., Li, S., Wang, F., Yang, Z., Zhang, W., Chen, S., & Yao, D. (2023).
A novel method for petroleum and natural gas resource potential evaluation and predic-
tion by support vector machines (SVM). Applied Energy, 351, 121836.
Yussuf, R. O., & Asfour, O. S. (2024). Applications of artificial intelligence for energy effi-
ciency throughout the building lifecycle: An overview. Energy and Buildings, 113903.
Zelenika, I., & Pearce, J. M. (2014). Innovation through collaboration: Scaling up solu-
tions for sustainable development. Environment, Development and Sustainability, 16,
1299–1316.
Zhang, W., Jiang, Y., & Zhang, W. (2019). Capabilities for collaborative innovation of tech-
nological alliance: A knowledge-based view. IEEE Transactions on Engineering
Management, 68(6), 1734–1744.
Zhuang, W., Chen, C., & Lyu, L. (2023). When foundation model meets federated learning:
Motivations, challenges, and future directions. arXiv preprint arXiv:2306.15546.
 9 Artificial Intelligence
Using Federated Learning
Manjushree Nayak and Debasish Padhi

INTRODUCTION
Artificial intelligence (AI) has transformed industries by enabling machines to learn
from data and make decisions autonomously. One innovative approach within AI is
federated learning, a technique that facilitates collaborative model training without
sharing raw data. This method addresses crucial issues like data privacy, minimiza-
tion, and access control, making it especially valuable in scenarios where central
data storage is impractical or poses privacy risks (Abadi et al., 2016).
Federated learning operates by training local models on diverse datasets held by
individual nodes, with only model parameters exchanged to create a global model.
Unlike traditional centralized machine learning, federated learning eliminates the
need to aggregate data in one location, thereby reducing privacy concerns associated
with data sharing. This decentralized training approach comes in various forms:
horizontal federated learning for similar datasets, vertical federated learning for
complementary datasets, and federated transfer learning for leveraging pre-trained
models.
The applications of federated learning are broad and impactful. In transportation,
autonomous vehicles utilize federated learning to improve safety by minimizing
data transfers and accelerating learning processes. Industries embracing Industry 4.0
benefit from federated learning’s privacy-preserving algorithms, ensuring data con-
fidentiality while enhancing operational efficiency. In healthcare, federated learning
enables collaborative AI model training across multiple medical institutions without
compromising patient privacy, leading to advancements in disease diagnosis and
treatment.
Federated learning’s capability to train AI models on decentralized data sources
is reshaping AI training methodologies. By processing data locally and tapping into
diverse information from various devices and sensors, federated learning offers a
secure and efficient means to leverage AI’s power without compromising privacy or
data security. As organizations prioritize data protection and regulatory compliance,
federated learning emerges as a crucial tool in advancing AI technologies across
sectors.
Federated learning represents a significant shift in AI training methods by
enabling collaborative model training while safeguarding data privacy and security.
Its applications across industries highlight its importance in driving innovation and
overcoming challenges associated with traditional centralized machine learning. As

156 DOI: 10.1201/9781003482000-9

Artificial Intelligence Using Federated Learning 157

AI evolves, federated learning stands out as a key enabler for unlocking AI’s full
potential in a privacy-conscious world.
The working of federated learning (FL) involves a decentralized and collab-
orative model training process:

1. Initialization: The central server establishes a global model, acting as the

initial foundation for the task at hand (McMahan et al., 2017a). This global
model serves as a starting point, incorporating foundational parameters that
guide subsequent learning processes. The central server then distributes
this initialized model to decentralized devices participating in FL. These
devices subsequently refine the model based on their local datasets.
2. Distribution: Instead of sending the entire model, which may contain sen-
sitive information, only model parameters or updates are transmitted to
individual client devices (McMahan et al., 2017a). This strategy minimizes
communication overhead and ensures data privacy. Each client device
receives these model fragments and refines the model locally based on its
dataset.
3. Training: Individual client devices autonomously refine the model using
their local datasets (McMahan et al., 2017a). This phase empowers each
client to learn from its unique data without compromising privacy, as raw
information remains decentralized. Local training allows models to adapt
to the specific characteristics of diverse datasets, capturing nuances and
patterns relevant to each device’s context.
4. Updates: Only refined model parameters, not raw data, are transmitted
back to the central server from individual client devices (McMahan et al.,
2017a). This privacy-preserving strategy ensures that sensitive information
remains localized, addressing concerns related to data security and privacy.
By sending only the model differentials or updates, FL optimizes com-
munication efficiency while allowing the central server to aggregate these
modifications and iteratively improve the global model.
5. Aggregate: The central server collects and integrates model updates from
individual client devices to enhance the global model (McMahan et al.,
2017a). During this phase, only refined model parameters are transmitted,
preserving the privacy of raw data. The central server utilizes aggrega-
tion algorithms to combine these updates effectively, considering factors
like weights or gradients, ensuring a comprehensive and improved global
model.
6. Iteration: Steps 2–5 are repeated iteratively for continuous model
improvement.
7. Convergence: This phase persists until the global model reaches a satisfac-
tory level of performance. Through successive rounds of model distribu-
tion, local training, and updates, the global model refines its parameters.
Convergence is achieved when the model reaches a point of stability or
optimization, reflecting a collective understanding from diverse datasets.
8. Deployment: The refined and converged model is made ready for inference
in real-world applications. This final model encapsulates collective insights
158 Artificial Intelligence Using Federated Learning

from decentralized client devices while upholding the principles of data

privacy and security, as raw data never leaves the local devices during the
learning process.

Considerations in federated learning include handling heterogeneity in data and

device capabilities, ensuring secure communication during transmission, addressing
potential communication delays, and tuning hyperparameters for optimal perfor-
mance. FL is especially valuable in privacy-focused applications like healthcare,
finance, and edge computing, where data sensitivity and resource constraints are
critical factors.
This chapter is organized into several sections to provide a comprehensive under-
standing of federated learning and its relationship with artificial intelligence.

LITERATURE REVIEW
Federated learning is a distributed machine learning approach that aims to preserve
privacy by keeping data locally while utilizing fragmented data and protecting cli-
ent privacy (Carl Smestad et al., 2023). FL was introduced in 2017 as a promising
framework for distributed machine learning that trains models without sharing local
data, thus ensuring user privacy by sharing unique data distribution properties (Attia
Qammar et al., 2022). The decentralized nature of FL allows clients to train inter-
mediate models on their devices with locally stored data, contributing to the col-
laborative building of a global model without compromising individual data privacy
(Liming Zhu et al., 2023).
One critical aspect of FL is the integration of blockchain technology to enhance
security and privacy. Blockchain has been identified as a solution to potential secu-
rity and privacy attacks in traditional federated learning, offering characteristics that
provide a secure environment for FL systems (Attia Qammar et al., 2022). By lever-
aging blockchain, FL systems can address challenges related to the disclosure of
private information, unreliable uploading of model parameters, and communication
costs, among others (Attia Qammar et al., 2022).
Client selection in federated learning is another crucial area of research. Random
client selection in FL can negatively impact learning performance due to various rea-
sons, leading researchers to explore client selection schemes that address challenges
such as heterogeneity, resource allocation, communication costs, and fairness (Carl
Smestad et al., 2023). Evaluating the impact of unsuccessful clients and gaining a
theoretical understanding of fairness in FL are highlighted as beneficial improve-
ments in client selection mechanisms (Jingyue Li et al., 2023).

NEED FOR FEDERATED LEARNING IN AI

The growth of Internet of Things (IoT) technologies and associated processes high-
lights the importance of reusing vast amounts of data (Smith et al., 2020a; Wang
et al., 2021). Advancements in big data analytics and computational methods, such as
machine learning and deep learning, have enabled users to effectively manage data.
Artificial intelligence applications are increasingly being used to address challenges
in optimized resource management, efficient antenna selection in wireless systems,
Artificial Intelligence Using Federated Learning 159

and dynamic communication network areas (Jones et al., 2019; Chen et al., 2022).
Traditional AI models often require users to share personal data with a central sys-
tem for learning purposes. A major concern with such approaches revolves around
the privacy of sensitive user data. Federated learning proves highly effective in sce-
narios where decision-making relies on extensive data distributed across multiple
training nodes, while also addressing privacy and security concerns (Li et al., 2021).
Machine learning models typically leverage data from various sources to make pre-
dictions. However, due to constraints such as bandwidth limitations, security consid-
erations, and storage capacity, transmitting raw data to a central location becomes
impractical. FL functions as a distributed learning model, preserving data privacy,
enabling efficient utilization of collected raw data, and transmitting it to a central
location. FL also plays a significant role in the development of smart cities, as dis-
cussed by researchers in recent studies (Zhang et al., 2023). Policymakers in smart
cities can utilize FL to transmit sensitive data collected from IoT devices for the
efficient operation of critical services. The FL framework allows users to access data
without compromising the privacy of others. Ultimately, the refined global model
generated by the system is distributed to all users, who then download the updated
global model and use local processing to enhance performance on their devices.

OVERVIEW
In federated learning, several styles and equations are employed to ease model train-
ing across distributed devices while conserving data sequestration.
Then are some commonly used methods and associated equations, along with an
extension of the handled information.

federated averaging
Federated averaging is a commonly used method in federated learning where local
model updates from participating devices are aggregated to update the global model.
The federated averaging can be expressed as:
1
E
N
0t +1 = wi 0i
N i =1

where:
0 t+1 represents the updated global model parameters.
N denotes the total number of participating devices.
0i signifies the model parameters of the ith device
wi indicates the weight assigned to the model update from the ith device

secUre aggregation
Secure aggregation ensures that model updates from participating devices are aggre-
gated in a privacy-preserving manner, safeguarding sensitive information. Secure
aggregation often utilizes cryptographic techniques such as homomorphic encryp-
tion or multiparty computation to aggregate encrypted model updates without reveal-
ing individual contributions.
160 Artificial Intelligence Using Federated Learning

differential Privacy
Differential privacy is a privacy-preserving mechanism that introduces noise to
individual data samples or model updates to prevent sensitive information leakage.
Adding Gaussian noise to model updates for achieving differential privacy can be
represented as:
~
0 i = 0i + N(0, s 2 )

where:
0~i denotes the perturbed model update.
0i represents the original model update from the ith device.
N(0,a2) signifies Gaussian noise with mean 0 and variance

federated learning witH deeP learning

Extending FL to deep neural network architectures enables collaborative model
training for tasks such as image classification and NLP. The equations used in fed-
erated learning with deep learning are similar to those in centralized deep learn-
ing, involving forward and backward propagation for model training but adapted for
decentralized and privacy-preserving FL.

Model oPtiMization tecHniqUes

Various optimization techniques like stochastic gradient descent (SGD) and adaptive
learning rate methods such as Adam are employed in FL to enhance model conver-
gence and performance. The equations for model optimization techniques in FL
resemble those in centralized machine learning, with adjustments made to accom-
modate the distributed nature of training and the necessity for privacy preservation.

CLASSIFICATION OF FEDERATED LEARNING

Federated learning can be categorized based on data distribution features. The data
matrix Di represents information from individual data owners, where each row and
column corresponds to a characteristic. Sample ID space (I), feature space (X), and
label space (Y) are considered. FL is classified as horizontally, vertically, or feder-
ated transfer learning (FTL) based on data dispersion among parties in the feature
and sample ID space. Schematic representations for a two-party situation include
federated transfer learning, vertical federated learning (VFL), and horizontal feder-
ated learning (HFL), as shown in Figure 9.1.

federated transfer learning

Federated transfer learning emerges as a solution when datasets not only vary in
sample size but also diverge in feature space (Chen et al., 2021; Pan and Yang, 2010).
For instance, consider a scenario where a Chinese bank and a U.S. e-commerce firm
Artificial Intelligence Using Federated Learning 161

FIGURE 9.1 Federated transfer learning.

share a small user population due to geographical constraints. FTL utilizes transfer
learning, a technique enabling the application of knowledge from one domain to
another, to generate solutions for the combined dataset with distinct features (Pan
and Yang, 2010). This approach goes beyond traditional federated learning methods,
addressing challenges posed by diverse feature spaces and ensuring more effective
collaboration between entities with disparate data distributions, ultimately enhanc-
ing the adaptability of AI models (Chen et al., 2021).

Vertical Federated Learning

Vertical federated learning is designed for scenarios where datasets share the same
sample ID space but diverge in feature space (McMahan et al., 2017b; Yang et al.,
2019a). Consider an e-commerce firm and a bank in the same city with distinct user
features. VFL, also known as feature-based FL, addresses this by aggregating dis-
parate features. It involves computing training loss and gradients collaboratively to
develop a joint model that leverages insights from both datasets (McMahan et al.,
2017b; Yang et al., 2019a). The security definition in VFL assumes participant hon-
esty and curiosity, highlighting the need for trust among entities. This approach
enhances collaboration between organizations with shared identifiers but differing
feature representations, fostering more comprehensive and effective machine-learn-
ing models (McMahan et al., 2017b; Yang et al., 2019a), as shown in Figure 9.2.

Horizontal Federated Learning

Horizontal federated learning is apt for datasets sharing a common feature space but
varying in sample size, as seen in two regional banks with similar business features
but diverse customer bases (Kairouz et al., 2019a; Smith et al., 2021). HFL employs
collaborative deep learning or limited parameter exchange to address challenges such
as stragglers and communication costs (Kairouz et al., 2019a; Smith et al., 2021). This
162 Artificial Intelligence Using Federated Learning

FIGURE 9.2 Vertical federated learning.

FIGURE 9.3 Horizontal federated learning.

approach facilitates joint model training while preserving data decentralization. The
security definition in HFL assumes an honest but curious server, acknowledging intru-
sion capability limited to hosted data participants (Kairouz et al., 2019a; Smith et al.,
2021). By accommodating variations in sample size while sharing feature spaces,
HFL emerges as a strategic solution, fostering effective collaboration and model
development across organizations with similar business, as shown in Figure 9.3.

APPLICATIONS
In sectors such as finance, sales, and various industries where data aggregation for
ML model training faces challenges like data security, privacy protection, and intel-
lectual property rights, federated learning emerges as a promising and innovative
Artificial Intelligence Using Federated Learning 163

modeling mechanism (Yang et al., 2019a; Li et al., 2020). Machine learning meth-
ods enable customer-specific services like product recommendations and sales
assistance by utilizing crucial data elements such as users’ buying power, personal
preferences, and product attributes (Nayak and Barman, 2022; Bishop, 2006; Hastie
et al., 2009). In practical applications, these data elements might be distributed
across different departments or companies. This situation presents two primary
challenges: overcoming data walls between entities like banks, social networking
sites, and e-commerce platforms, and addressing the issues of traditional, heteroge-
neous data that cannot be easily handled by conventional ML approaches (Caruana,
1997; Domingos, 2012).
Standard machine learning methods struggle to resolve these challenges, hin-
dering the broader acceptance and utilization of AI. Federated and transferable
learning offer solutions to these issues. Federated learning facilitates model con-
struction for all involved parties without the need to aggregate company data,
ensuring data security and privacy and providing personalized services to cus-
tomers (McMahan et al., 2017a; Li et al., 2020). Simultaneously, transfer learning
addresses data heterogeneity and overcomes the limitations of conventional AI
approaches (Pan and Yang, 2010; Weiss et al., 2016). This approach allows the
creation of a cross-company, cross-data, and cross-domain big data and artificial
intelligence ecosystem using federated learning. The federated learning architec-
ture enables querying multiple databases without exposing any data, enhancing
privacy (Yang et al., 2019a).
In the financial sector, federated learning can address risks associated with mul-
tiparty borrowing, a historical source of risk in banking, by identifying individu-
als involved in such activities without disclosing user lists between banks (Huang
et al., 2021). Encrypted federated learning methods can secure sensitive information
while revealing critical insights. In the realm of smart healthcare, federated learning
holds great potential (Liang et al., 2020). Despite the confidentiality and sensitiv-
ity of medical datasets, their isolated existence in various medical institutions and
hospitals poses a challenge for comprehensive data gathering. Federated learning,
combined with transfer learning, emerges as an effective strategy to enhance ML
algorithm efficiency by allowing medical institutions to collaborate, share data, and
fill gaps in information (Sheller et al., 2020). Federated transfer learning becomes
crucial for advancing smart healthcare and elevating human healthcare to new levels
of performance (Rajkomar et al., 2018).

CHALLENGES AND FUTURE OF FEDERATED LEARNING

Challenges
Privacy
Privacy concerns in federated learning encompass data privacy, where models train
on decentralized data without directly sharing raw information. Safeguarding sensi-
tive details during training poses challenges despite not centralizing data. Model
updates also present risks, as transmitting updates may inadvertently expose sensi-
tive information (Bonawitz et al., 2017). Robust encryption and privacy-preserving
164 Artificial Intelligence Using Federated Learning

methods are imperative to mitigate these risks, ensuring that refined model param-
eters do not compromise confidentiality (Shokri et al., 2015). Addressing privacy
issues in FL requires a delicate balance between collaborative model training and
protecting individual data, necessitating ongoing advancements in encryption tech-
niques and privacy-preserving technologies to foster trust and security in decentral-
ized machine learning paradigms (Nayak and Narain, 2020a).

Security
Security challenges in federated learning include model poisoning, where malicious
participants may inject poisoned data or model updates, compromising the global
model’s integrity (Bagdasaryan et al., 2018). This threat underscores the importance
of robust model validation mechanisms. Communication security poses risks during
model update transmission between clients and the central server (Hardy et al., 2017).
Protecting against eavesdropping and maintaining communication confidentiality is
complex, requiring encryption and secure communication protocols. Implementing
measures to detect and mitigate model poisoning, coupled with stringent communi-
cation security protocols, is essential for fortifying FL against adversarial attacks,
ensuring the trustworthiness of the collaborative learning process and the integrity
of the resultant global model.

Diversity of Data
In federated learning, diversity in data manifests through variations in data distri-
bution, quality, and characteristics across different client devices (Kairouz et al.,
2019a). Managing the complexities arising from diverse datasets poses a challenge,
as models must exhibit robustness across heterogeneous data sources. The challenge
lies in reconciling disparities in the types, scales, and representations of data among
clients. Strategies to ensure model resilience involve techniques such as weighted
aggregation to account for varying data contributions (Li et al., 2020). Addressing
this diversity is pivotal for FL’s success, requiring continuous advancements in algo-
rithmic approaches that can adapt and generalize effectively amidst the inherent
heterogeneity within the decentralized learning framework.

Communication Overlap
Communication overhead is a critical challenge in federated learning, encompass-
ing network latency and bandwidth constraints (Chen et al., 2021). Network latency
introduces delays in communication between the central server and client devices,
impacting the efficiency of model updates. This challenge is particularly pronounced
in real-time applications where timely responses are crucial. Bandwidth constraints
pose another issue, especially when handling large model updates or extensive
datasets (Bonawitz et al., 2017). Transmitting such data strains network bandwidth,
becoming especially problematic for resource-constrained devices. Mitigating these
challenges requires optimizing communication protocols, exploring edge comput-
ing solutions, and implementing compression techniques to minimize the impact
of communication overhead on the overall performance of the federated learning
process.
Artificial Intelligence Using Federated Learning 165

Model Aggregation and Quality

Model aggregation and quality in federated learning present challenges in both
aggregation methods and model divergence. An effective combination of model
updates demands robust aggregation techniques, as simple averaging may fall short
in addressing variations in local data. Sophisticated methods are essential to recon-
cile disparities among client contributions and prevent biased aggregation. Model
divergence is a significant challenge, necessitating accurate convergence of the
global model while accommodating differences in local datasets (McMahan et al.,
2017a). Ensuring harmonious integration of diverse insights without compromising
overall model accuracy requires continuous refinement of aggregation strategies and
convergence mechanisms, highlighting the complexities inherent in achieving high-
quality models in decentralized learning environments.

Regulatory and Legal Adherence

Regulatory and legal adherence in federated learning involves challenges related to
data ownership and legal frameworks (Chen et al., 2021). Determining ownership
and control of models trained on decentralized data presents intricacies, demanding
careful consideration for regulatory compliance, especially concerning data gover-
nance and ownership (Hardy et al., 2017). Adhering to privacy regulations and legal
frameworks across diverse regions or jurisdictions becomes crucial in FL scenarios
involving participants from various locations (Kairouz et al., 2019a). Navigating
these complexities requires a nuanced understanding of international data protection
laws, transparency in ownership agreements, and the development of standardized
practices that align with the evolving regulatory landscape, ensuring ethical and law-
ful practices in the collaborative and decentralized context of FL.

Resource Limitations
Resource limitations in federated learning are notably pronounced in edge device.
Extending FL to these devices, characterized by limited computational resources,
introduces challenges in optimizing model updates and learning processes. The con-
strained nature of edge devices necessitates the development of lightweight models,
efficient communication protocols, and strategies to mitigate computational burdens
during local training (Li et al., 2019). Balancing the collaborative learning objec-
tives with the resource constraints of edge devices requires innovative solutions such
as edge-friendly algorithms, compression techniques, and federated optimization
approaches tailored for low-power and resource-constrained environments, ensuring
the feasibility and effectiveness of federated learning on the edge.

Scalability
Scalability in federated learning confronts challenges, particularly when expanding
to a larger participant base (Smith et al., 2021). As the number of clients increases,
coordinating model updates and efficiently aggregating information becomes intri-
cate. The sheer scale introduces communication overhead, potentially impacting
the responsiveness and efficiency of the learning process. Effective coordination
demands scalable algorithms, optimized communication protocols, and strategies
166 Artificial Intelligence Using Federated Learning

to mitigate potential bottlenecks (Chen et al., 2021). Innovations in federated opti-

mization techniques, decentralized aggregation methods, and distributed learning
approaches are essential for ensuring the scalability of FL systems. Addressing these
challenges is pivotal to unleashing the full potential of federated learning across
diverse applications and accommodating the complexities of large-scale, decentral-
ized collaboration.

Future
The future of federated learning in the realm of artificial intelligence holds signifi-
cant promise and potential for revolutionizing collaborative machine learning while
preserving data privacy. Here are key insights from the provided sources regarding
the future of federated learning:

Enhanced Privacy Protection

Federated learning allows data to be processed locally on devices, maintaining pri-
vacy by keeping sensitive data on the device and only sending model updates to a
central server for processing. This approach ensures data security, especially for
sensitive information like medical records or financial data.

Collaborative Model Training

FL enables collaborative model training on distributed data sources, allowing mod-
els to be trained on data from multiple devices or locations, leading to more accurate
and diverse models. This collaborative aspect is crucial for fields like healthcare,
where diverse data sources can enhance model representativeness.

Overcoming Data Limitations

For companies with small datasets, federated machine learning (FedML) offers a
solution by enabling small-data organizations to train sophisticated machine learn-
ing models through decentralized data sharing. This approach helps bridge the
digital divide between companies with vast data resources and those with limited
datasets.

Incorporating New Knowledge

The concept of federated learning with new knowledge focuses on effectively inte-
grating various new knowledge sources into existing FL systems to reduce costs,
extend system lifespan, and promote sustainable development. This highlights the
adaptability and evolution of FL systems to incorporate emerging knowledge.

Future Applications
The future of federated learning is poised to witness a surge in new applications
leveraging FL, enhancing user experiences in unprecedented ways. Applications like
self-driving connected cars can utilize FL to make safer decisions by leveraging col-
lective data from similar scenarios.
Artificial Intelligence Using Federated Learning 167

CASE STUDIES
Healthcare
In healthcare, FL empowers collaborative efforts among hospitals to enhance pre-
dictive models for disease diagnosis and treatment recommendations (Rieke et al.,
2020). Utilizing this approach, institutions can construct models predicting patient
outcomes and also tailoring treatment plans. Crucially, federated learning gives
safeguards to patient privacy by allowing institutions to share insights without
compromising sensitive data (Sheller et al., 2020). This privacy-compliant method-
ology optimizes healthcare decision-making, fostering a collective intelligence that
improves overall patient care, as shown in Figure 9.4.

Smart Cities
In smart city development, federated learning enables collaborative initiatives among
diverse municipal agencies to elevate services and infrastructure through AI (Dinh
et al., 2019). This methodology, when applied, also refines urban functionalities like
traffic management, public transportation scheduling, and energy consumption pre-
dictions. Federated learning ensures the preservation of data privacy for each agency
involved, fostering a secure and efficient exchange of insights (Zhao et al., 2021). By
optimizing these critical aspects of urban living, federated learning empowers cities

FIGURE 9.4 Application of FL in healthcare.

168 Artificial Intelligence Using Federated Learning

FIGURE 9.5 Application of FL in smart cities.

to evolve intelligently, creating a sustainable and interconnected environment while

upholding the confidentiality of sensitive municipal data as shown in Figure 9.5.

Finance
In the financial sector, federated learning facilitates collaborative initiatives as dis-
parate banks and financial institutions unite to fortify fraud detection models
(Bortolameotti et al., 2019). By employing federated learning, a potent fraud detection
system evolves, drawing insights from diverse sources without compromising the con-
fidentiality of sensitive customer information (Yang et al., 2020b). This decentralized
approach allows the amalgamation of data without the necessity of sharing individual
client details, ensuring stringent privacy standards. Consequently, this federated learn-
ing methodology enhances the collective resilience of fraud detection systems across
the financial landscape, fortifying the industry against evolving threats while preserv-
ing the integrity of client data within each contributing institution as shown in figure.

Internet of Things
In the realm of edge devices and IoT, federated learning fosters collaboration among
smart devices, including smartphones and IoT devices, to amplify local AI capa-
bilities (Kairouz et al., 2019a). This innovative approach allows for the enhance-
ment of speech recognition, image processing, and other AI applications directly on
edge devices. Notably, federated learning achieves this without the need to transmit
sensitive data to a centralized server, preserving user privacy and security (Yang
et al., 2018). By distributing the learning process across decentralized devices, feder-
ated learning optimizes the efficiency of local AI applications, ensuring a seamless
user experience while upholding the confidentiality of individual data in the rapidly
expanding landscape of interconnected devices, as shown in Figure 9.7.
Artificial Intelligence Using Federated Learning 169

FIGURE 9.6 Application of FL in finance.

FIGURE 9.7 Application of FL in IoT.

170 Artificial Intelligence Using Federated Learning

Education
In education, federated learning facilitates collaboration among diverse educational
institutions aiming to craft personalized learning models for students (Smith et al.,
2020a). This innovative approach utilizes federated learning to construct models that
dynamically adapt to individual learning styles. Crucially, this adaptation occurs
without compromising the privacy of student data (Hard et al., 2021). By distribut-
ing the learning process across institutions and respecting data privacy, federated
learning ensures the creation of effective and tailored educational experiences. This
collaborative methodology not only enhances the quality of personalized learning
but also upholds the confidentiality of student information, offering a cutting-edge
solution to meet the evolving needs of diverse learners in an interconnected educa-
tional landscape.

FIGURE 9.8 Application of FL in telecommunication.

Artificial Intelligence Using Federated Learning 171

Telecommunications
Within the telecommunications sector, federated learning revolutionizes collabora-
tion as telecom companies join forces to refine network optimization and predictive
maintenance models (Abdelkhalek et al., 2020). Employing federated learning, this
scenario enables a heightened accuracy in predicting network failures and optimiz-
ing data traffic across diverse regions (Jiang et al., 2021). A crucial aspect is that fed-
erated learning achieves these improvements without jeopardizing user data privacy.
By decentralizing the learning process, telecom companies can collectively enhance
the reliability and efficiency of their networks, ensuring a seamless user experience
while upholding the stringent standards of privacy and security. This collaborative
approach marks a transformative leap in the evolution of telecommunications infra-
structure and services.

CONCLUSION
Federated learning stands as a transformative paradigm within the realm of artificial
intelligence, fostering collaborative innovation across industries while simultane-
ously addressing the paramount concerns of privacy and data security (McMahan
et al., 2017a). This decentralized learning approach represents a revolutionary force,
allowing multiple entities to collectively enhance models without the necessity of
sharing raw data. This unique approach strikes a nuanced balance between leverag-
ing collective intelligence for advancement and preserving individual privacy rights.
Collaboratively refining machinery maintenance processes without compromising
sensitive data allows industries to minimize downtime and improve overall equip-
ment efficiency. This translates into a more sustainable and cost-effective approach
to manufacturing, highlighting the transformative influence of federated learning on
industrial processes. Edge devices and the Internet of Things also witness substantial
benefits from federated learning (Kairouz et al., 2019a). By distributing the learning
process across devices, federated learning optimizes local AI capabilities without the
need to transmit sensitive data to centralized servers. This ensures efficient speech
recognition, image processing, and other AI applications directly on edge devices,
contributing to a seamless user experience while respecting user privacy, a crucial
factor in the era of interconnected devices. Telecommunications undergo a trans-
formative leap with the implementation of federated learning (Abdelkhalek et al.,
2020). Telecom companies collaboratively improve network optimization and pre-
dictive maintenance models, enhancing the accuracy of predicting network failures
and optimizing data traffic across diverse regions (Nayak et al., 2023). Crucially, this
is achieved without compromising user data privacy, resulting in a more reliable and
efficient telecommunications infrastructure that caters to the evolving needs of users
while upholding stringent standards of privacy and security.
Federated learning not only propels the development of more accurate and effi-
cient models but also underscores the significance of respecting individual privacy
in the pursuit of collective intelligence (Yang et al., 2019a). As we navigate the future
of AI, federated learning represents a pivotal step towards a more collaborative,
secure, and ethically grounded approach to artificial intelligence, shaping a future
172 Artificial Intelligence Using Federated Learning

where technological advancement coexists harmoniously with privacy and ethical

considerations.

REFERENCES
Abadi, M., et al. (2016). “TensorFlow A System for Largescale Machine Knowledge.” In
Proceedings of the 12th USENIX Conference on Operating Systems Design and
Performance (OSDI). USENIX Association.
Abdelkhalek, N., et al. (2020). “Federated Learning in Industrial Wireless Networks
an Overview, Taxonomy, and Disquisition Directions.” IEEE Deals on Industrial
Informatics, 16(3), 1802–1810.
Bagdasaryan, E., et al. (2018). “How to Backdoor Federated Learning.” arXiv preprint
arXiv1807.00459.
Bishop, C.M. (2006). Pattern Recognition and Machine Knowledge. Springer.
Bonawitz, K., et al. (2017). “Practical Secure Aggregation for Federated Learning on User-
Held Data.” arXiv preprint arXiv1611.04482.
Bortolameotti, R., et al. (2019). “Federated Learning for Insulation: Conserving IoT
Analytics.” IEEE Internet of Goods Journal, 6(2), 3702–3713.
Caruana, R. (1997). “Multitask Learning.” Machine Knowledge, 28(1), 41–75.
Chen, C.Z., Liao, W., Hua, K., Lu, C., & Yu, W. (2021). “Towards Asynchronous
Federated Learning for Heterogeneous Edge-Powered Internet of Things.” Digital
Communications and Networks, 7(3), 317–326. ISSN 2352-8648, https://doi.
org/10.1016/j.dcan.2021.04.001. (https://www.sciencedirect.com/science/article/pii/
S2352864821000195)n
Chen, D., et al. (2022). “Deep Knowledge-Predicated Resource Operation in 6G and Beyond
a Comprehensive Survey.” IEEE Network, 36(1), 136–142.
Chen, M., et al. (2021). “Federated Transfer Learning for Cross: Sphere Recommendation.”
IEEE Deals on Neural Networks and Learning Systems, 32(3), 921–934.
Dinh, H.T., et al. (2019). “Federated Learning Challenges, Styles, and Future Directions.”
IEEE Deals on Arising Motifs in Computational Intelligence, 4(3), 288–298.
Domingos, P. (2012). “A numerous Useful Goods to Know About Machine Learning.”
Dispatches of the ACM, 55(10), 78–87.
Hard, A., et al. (2021). “Federated Learning Challenges, Styles, and Future Directions.” IEEE
Deals on Arising Motifs in Computational Intelligence, 5(4), 264–279.
Hardy, S., Henecka, W., Ivey-Law, H., Nock, R., Patrini, G., Smith, G., & Thorne, B.
(2017). Private Federated Learning on Vertically Partitioned Data via Entity
Resolution and Additively Homomorphic Encryption. https://api.semanticscholar.org/
CorpusID:26233593
Hastie, T., et al. (2009). The Rudiments of Statistical Learning Data Mining, Inference, and
Prophecy. Springer.
Huang, Y., et al. (2021). “Secure Federated Learning with an Untrusted Central Garçon.”
arXiv preprint arXiv2102.05580.
Jiang, B., et al. (2021). “Edge Intelligence in 6G Openings and Challenges.” IEEE Network,
35(1), 24–31.
Jones, C., et al. (2019). “Machine Knowledge for Wireless Communication Systems a
Tutorial.” IEEE Dispatches Checks & Tutorials, 21(4), 3039–3071.
Kairouz, P., et al. (2019a). “Advances and Open Problems in Confederated Knowledge.” arXiv
preprint arXiv1912.04977.
Artificial Intelligence Using Federated Learning 173

Li, J., Meland, P.H., Notland, J.S., Storhaug, A., & Tysse, J.H. (2023). “Evaluating the
Impact of ChatGPT on Exercises of a Software Security Course.” 2023 ACM/IEEE
International Symposium on Empirical Software Engineering and Measurement
(ESEM), New Orleans, LA, USA, pp. 1–6, doi: 10.1109/ESEM56168.2023.10304857.
Li, W., et al. (2019). “Insulation: Conserving Federated Learning for Internet of Goods a
Survey.” IEEE Access, 7, 166156–166173.
Li, X., et al. (2020). “Federated Learning for Insulation: Conserving Healthcare Applications
a Survey.” IEEE Deals on Industrial Informatics, 17(6), 4139–4146.
Li, X., et al. (2021). “Federated Learning for Insulation: Conserving Healthcare Applications
a Survey.” IEEE Deals on Industrial Informatics, 17(6), 4139–4146.
Liang, S., et al. (2020). “Federated Learning for Healthcare a Review.” IEEE Access, 8,
180916–180936.
McMahan, H.B., et al. (2017a). “Communication-Effective Knowledge of Deep Networks
from Decentralized Data.” In Proceedings of the 20th International Conference on
Artificial Intelligence and Statistics (AISTATS), Fort Lauderdale, FL, USA. JMLR:
W&CPvolume 54. Copyright 2017. 1602.05629 (arxiv.org)
McMahan, H.B., et al. (2017b). “Communication-Effective Knowledge of Deep Networks
from Decentralized Data.” arXiv preprint arXiv1602.05629.
Nayak, M., & Barman, A. (2022). “A Real-Time Pall-Predicated Healthcare Monitoring
System.” In S. Kshatri, K. Thakur, M. Khan, D. Singh, & G. Sinha (Eds.), Computational
Intelligence and Applications for Afflictions and Healthcare (pp. 229–247). IGI Global.
https//doi.org/10.4018/978-1-7998-9831-3.ch011
Nayak, M., Dass, A.K., & Kshatri, S.S. (2023). “An AI-Predicated Effective Model for the
Type of Business Signals Using Convolutional Neural Network.” In S. Dewangan, S.
Kshatri, A. Bhanot, & M. Shah (Eds.), Building Secure Business Models Through
Blockchain Technology Tactics, Styles, Limitations, and Performance (pp. 20–35). IGI
Global. https//doi.org/10.4018/978-1-6684-7808-0.ch002
Nayak, M., & Narain, B. (2020a). “Big Data Mining Algorithms for Predicting Dynamic
Product Price by Online Analysis.” In Computational Intelligence in Data Mining
Proceedings of the International Conference on ICCIDM 2018 (pp. 701–708).
Springer.
Nayak, M., & Narain, B. (2020b). “Predicting Dynamic Product Price by Online Analysis
Modified K-Means Cluster.” In A. Das, J. Nayak, B. Naik, S. Dutta, & D. Pelusi (Eds.),
Computational Intelligence in Pattern Recognition. Advances in Intelligent Systems
and Computing, vol. 1120. Springer. https//doi.org/10.1007/978-981-15-2449-3_1
Pan, S.J., & Yang, Q. (2010). “A Check on Transfer Knowledge.” IEEE Deals on Knowledge
and Data Engineering, 22(10), 1345–1359.
Qammar, A., Ding, J., & Ning, H. (2022). “Federated Learning Attack Surface: Taxonomy,
Cyber Defences, Challenges, and Future Directions”. Artificial Intelligence Review, 55,
3569–3606. https://doi.org/10.1007/s10462-021-10098-w
Rajkomar, A., et al. (2018). “Scalable and Accurate Deep knowledge with Electronic Health
Records.” NPJ Digital Medicine, 1(1), 1–10.
Rieke, N., et al. (2020). “The Future of Digital Health with Federated Learning.” NPJ Digital
Medicine, 3(1), 1–6.
Sheller, M.J., et al. (2020). “Federated Learning in Medicine FacilitatingMulti- Institutional
Collaborations Without Sharing Patient Data.” Scientific Reports, 10(1), 12598.
Shokri, R., et al. (2015). “Insulation-Conserving Deep Knowledge.” In Proceedings of the
22nd ACM SIGSAC Conference on Computer and Dispatches Security (CCS), ACM
Conferences, New York, NY. Association for Computing Machinery.
174 Artificial Intelligence Using Federated Learning

Smestad, C., & Li, J. (2023). A Systematic Literature Review on Client Selection in
Federated Learning. Association for Computing Machinery. https://doi.org/10.1145/
3593434.3593438, ISBN: 9798400700446
Smith, A., et al. (2020a). “IoT Analytics Openings, Challenges, and Operations.” IEEE
Internet of Goods Journal, 7(12), 11518–11528.
Smith, A., et al. (2021). “Federated Learning Strategies, Challenges, and Future Directions.”
ACM Computing Checks, 54(6), Composition 118.
Wang, B., et al. (2021). “Towards Effective IoT Data Analytics a Comprehensive Review.”
IEEE Deals on Industrial Informatics, 17(6), 3845–3854.
Weiss, K., et al. (2016). “Check of Transfer Knowledge.” Journal of Machine Learning
Research, 17(1), 1–104.
Yang, Q., et al. (2018). “Federated Machine Learning Concept and Applications.” ACM Deals
on Intelligent Systems and Technology (TIST), 10(2), 1–19.
Yang, Q., et al. (2019a). “Federated Learning.” Emulsion Lectures on Artificial Intelligence
and Machine Learning, 13(3), 1–207.
Yang, Q., et al. (2020b). “Federated Machine Learning Concept and Applications.” ACM
Deals on Intelligent Systems and Technology (TIST), 10(2), 1–19.
Zhang, Y., et al. (2023). “Advancing Smart Cosmopolises Through Federated Learning
Openings and Challenges.” IEEE Deals on Sustainable Computing, 8(1), 114–127.
Zhao, L., et al. (2021). “Secure Federated Learning for Smart Metropolises Challenges and
Openings.” IEEE Internet of Goods Journal, 8(15), 12248–12257.
Zhu, L., Xu, X., Lu, Q., Governatori, G., & Whittle, J. (2022). AI and Ethics—Operationalizing
Responsible AI. In F. Chen & J. Zhou (Eds.), Humanity Driven AI. Springer. https://doi.
org/10.1007/978-3-030-72188-6_2
10 Applications of Federated
Learning in AI, IoT,
Healthcare, Finance,
Banking, and Cross-
Domain Learning
Walaa Hassan and Habiba Mohamed

Abbreviations
AI Artificial Intelligence
FL Federated Learning
IoT Internet of Things
HIPAA Health Insurance Portability and Accountability Act
FedAvg Federated Averaging
SMC Secure Multiparty Computation
SIEM Security Information and Event Management
IDS Intrusion Detection Systems
APTs Advanced Persistent Threats
API Application Programming Interface
ML Machine Learning
DL Deep Learning
VR Virtual Reality
SVM Support Vector Machine
CNN Convolutional Neural Network
NLP Natural Language Processing
GDPR General Data Protection Regulation
FLEG Federated Learning with Encrypted Gradients

INTRODUCTION
A new approach to training machine learning models known as federated learning
is becoming very popular. Instead of collecting all the data in one place, it distrib-
utes the work of improving models between many devices or servers. This helps
keep people’s data private and safe. It also means data transmission is unnecessary
from device to device. Federated learning allows different devices to work together

DOI: 10.1201/9781003482000-10 175

176 Artificial Intelligence Using Federated Learning

to make models better without sharing private information. This big change helps
artificial intelligence grow without risking privacy. Federated learning can be used
in many different fields. It is flexible and can help technology advance in many areas.
In the domain of the Internet of Things (IoT), federated learning enables edge
devices to collaboratively learn from locally generated data while preserving user
privacy, thereby enhancing the efficiency of IoT systems without compromising data
security [1]. Healthcare stands as another domain ripe for the integration of feder-
ated learning, where sensitive patient data can be kept within hospital or device
boundaries while still contributing to the improvement of healthcare algorithms [2].
Furthermore, federated learning holds promise in finance and banking, where insti-
tutions can collaborate on model training while adhering to strict regulatory frame-
works governing data privacy and security [3].
Moreover, the concept of federated learning extends beyond individual domains,
paving the way for cross-domain learning where models trained on data from one
domain can undergo transfer and adapted to another domain. This cross-domain
applicability opens up avenues for synergistic collaborations and knowledge transfer
between different industries, fostering innovation and driving progress on a broader
scale [4].
In addition to its applications in specific domains, federated learning also
addresses broader challenges in AI, such as data heterogeneity and scalability.
By allowing models to be trained across distributed datasets without centralized
aggregation, federated learning accommodates the diverse and often fragmented
nature of data sources, enhancing the resilience and capacity for generalization of
AI systems [5]. Furthermore, federated learning facilitates ongoing learning and
adaptation within dynamic environments, making it particularly well-suited for
scenarios where data distributions evolve, such as in online learning and real-time
analytics [6].
The potential of federated learning extends beyond traditional machine learning
tasks to encompass a broad array of AI applications, encompassing natural language
processing, as well as computer vision, and recommendation systems. In natural lan-
guage processing, federated learning enables the collaborative training of language
models across multiple organizations or jurisdictions while respecting privacy and
data sovereignty concerns [7]. Similarly, federated learning enhances the develop-
ment of computer vision algorithms by leveraging diverse datasets from edge devices
or distributed servers, leading to more robust and inclusive models [8]. Moreover,
federated learning empowers personalized recommendation systems by harnessing
user data from various sources without compromising individual privacy, thereby
improving the relevance and accuracy of recommendations [9].
Furthermore, the integration of federated learning techniques in edge comput-
ing environments holds significant promise for enhancing the efficiency and scal-
ability of AI systems. By leveraging the computational resources available on edge
devices, federated learning enables on-device model training and inference, reduc-
ing latency and bandwidth requirements associated with centralized approaches [10].
This decentralized approach also mitigates privacy concerns by keeping sensitive
data local to the device, thereby fostering user trust and compliance with privacy
regulations [11].
Applications of Federated Learning 177

In the context of autonomous vehicles and smart transportation systems, feder-

ated learning facilitates the collaborative training of AI models across vehicles and
infrastructure components, enabling real-time updates and improvements to percep-
tion and decision-making algorithms [12]. Similarly, in smart grid systems, federated
learning enables distributed optimization and control of energy resources preserving
the privacy of individual consumers [13]. These applications underscore the transfor-
mative potential of federated learning in shaping the future of AI-driven technolo-
gies across diverse domains and industries.
Moreover, federated learning holds promise in addressing challenges related to
data privacy and regulatory compliance in industries such as pharmaceuticals and
biotechnology. By allowing pharmaceutical companies to collaborate on model train-
ing while keeping sensitive patient data decentralized, federated learning facilitates
the development of personalized medicine and drug discovery algorithms without
compromising patient privacy [14]. Similarly, in the biotechnology sector, federated
learning enables research institutions to pool genomic data from diverse sources for
analysis and model training while adhering to data protection regulations and ethical
guidelines [15].
Furthermore, federated learning is poised to revolutionize the field of environ-
mental monitoring and conservation by enabling the collaborative analysis of sensor
data from various sources, including satellites, drones, and IoT devices. By leverag-
ing federated learning techniques, environmental scientists can develop AI models to
predict climate patterns, monitor wildlife populations, and assess ecosystem health
while preserving the privacy of sensitive location data and biodiversity records [16].
The applications of federated learning are not limited to specific industries but
extend to various societal challenges, including disaster response and humanitarian
aid. By enabling the collaborative analysis of data from multiple sources, including
social media, remote sensing, and government agencies, federated learning empow-
ers disaster response teams to make informed decisions in real-time, optimizing
resource allocation and emergency relief efforts [17].
In summary, federated learning represents a groundbreaking paradigm shift in
machine learning, offering a privacy-preserving and collaborative approach to model
training that transcends traditional centralized methods. With its diverse applica-
tions spanning across industries and domains, federated learning is poised to drive
innovation, empower communities, and shape the future of AI-driven technologies
in a decentralized and inclusive manner.
The chapter is organized as follows: In the second section, the most recent
approaches used in the area of FL are introduced. The concepts and frameworks
of FL are explained in the third section. Federated learning applications in various
domains are discussed in the fourth section. The fifth section explains the prospects
for federated learning and challenges, and the last part is a brief conclusion of the
contents of the chapter.

LITERATURE REVIEW
Federated learning, a cutting-edge approach in the field of machine learning, has
emerged as a transformative paradigm that addresses key challenges associated with
178 Artificial Intelligence Using Federated Learning

traditional centralized training methods. Unlike conventional methods where data is

consolidated and stored within a central server for model training, federated learn-
ing facilitates collaborative model training across numerous decentralized devices
or servers while keeping the raw data local and private [5]. This decentralized nature
of federated learning provides several benefits, such as improved data privacy and
security, reduced communication overhead, and scalability to large and heteroge-
neous datasets [10].
At its core, federated learning operates on the principle of collaborative model
training, where each participating device or server independently computes model
updates based on its local data and shares only the model parameters with a central
coordinator [4]. The central coordinator aggregates these parameters to update the
global model, which is then redistributed to the participating devices for further
refinement. This iterative process of local model training and global aggregation
continues until the desired level of model performance is achieved [18].
One of the key benefits of federated learning is its ability to preserve data privacy
while allowing for model training on sensitive or proprietary datasets. By keeping
data local to each device or server, federated learning minimizes the risk of data
exposure or unauthorized access, making it particularly well-suited for applications
in healthcare, finance, and other regulated industries [3]. Moreover, federated learn-
ing enables efficient utilization of distributed computing resources, allowing for
model training on edge devices, IoT devices, or cloud servers without the need for
centralized data aggregation [1].
In recent years, federated learning has gained significant traction across various
domains, with re-researchers and practitioners exploring its applications in fields
such as healthcare informatics, autonomous systems, and personalized recommen-
dation systems [19]. The versatility of federated learning stems from its ability to
adapt to diverse data distributions and privacy requirements, making it applicable to
a wide range of use cases and scenarios [8].
Federated learning’s decentralized approach not only addresses privacy concerns
but also enhances scalability, particularly in scenarios where data sources are dis-
tributed across geographically dispersed locations or edge devices. This distributed
nature of federated learning allows for the utilization of local data without the need
for data aggregation, reducing communication costs and latency associated with cen-
tralized approaches [11]. Moreover, federated learning accommodates data heteroge-
neity by enabling model training on diverse datasets while preserving the integrity
and privacy of each source [20].
The concept of federated learning has gained significant momentum in recent
years, driven by advancements in communication technologies, distributed comput-
ing, and privacy-preserving techniques. Researchers and practitioners are actively
exploring new methodologies and frameworks to overcome challenges such as
model synchronization, data skewness, and security vulnerabilities inherent in
federated learning systems [21]. Additionally, federated learning holds promise for
applications in dynamic and resource-constrained environments, such as IoT net-
works, where data privacy, bandwidth limitations, and energy efficiency are critical
considerations [22].
Applications of Federated Learning 179

Furthermore, federated learning extends beyond traditional machine learning

tasks to encompass federated reinforcement learning, federated transfer learning,
and other advanced techniques that leverage collaborative model training across
distributed data sources [13]. These advancements open up new possibilities for
federated learning in domains such as robotics, autonomous systems, and per-
sonalized AI assistants, where real-time adaptation and personalization are para-
mount [23].
The paper by Ghadi et al. [24] explores the challenges and potential solutions
for integrating federated learning (FL) with the Internet of Things. The authors
highlight the advantages of FL, which include safeguarding user privacy,
enhancing model performance, enabling scalable adaptability, and improving
learning quality within IoT networks. The study highlights various hurdles
faced by FL in IoT contexts, like resource management, privacy protection,
security issues, communication challenges, standardization concerns, and the
implementation of machine learning capabilities on IoT devices. The article not
only identifies these key challenges but also outlines potential areas for future
research. The literature review section discusses common FL-IoT applications,
while the methods and techniques section delves into the intricacies of FL and
IoT challenges. The results and discussion section offers insights into solutions
and opportunities for addressing FL-IoT challenges. The results and discussion
section offers insights into solutions and opportunities for addressing FL-IoT
challenges. Overall, the paper provides a comprehensive overview of the current
landscape of FL integration with IoT and sets the stage for further exploration
in this evolving field.
Federated learning is becoming increasingly popular in healthcare as a solu-
tion to the challenges of creating accurate machine learning models while protect-
ing patient privacy. This paper [25] enables model training without the need to
share sensitive patient data across different organizations. By leveraging data from
various healthcare institutions, FL offers a way to build robust AI models while
maintaining data security and confidentiality. The study highlights the importance
of effective communication, standardized data preparation, and regulatory compli-
ance for the successful implementation of FL models in healthcare. Current appli-
cations of FL in ophthalmology demonstrate its potential to transform medical
device development by utilizing diverse data sets while ensuring patient privacy.
Collaboration among stakeholders, including developers, healthcare providers,
and regulatory bodies, is crucial to address challenges like data labeling inconsis-
tencies and privacy attacks, ensuring the ethical and efficient use of FL-enabled
devices in healthcare settings.
Table 10.1 shows the federated learning domains and more of related studies.
Federated learning represents a pivotal advancement in machine learning that
addresses the challenges of data privacy, scalability, and heterogeneity in distributed
environments. With its decentralized and collaborative framework, federated learn-
ing is poised to revolutionize various industries and domains, driving innovation and
empowering organizations to leverage the collective intelligence of distributed data
sources while respecting individual privacy rights.
180 Artificial Intelligence Using Federated Learning

TABLE 10.1
Advantages, Disadvantages, and Techniques of Federated Learning Research
Reference Advantages Disadvantages Domains
[5] Provides a foundational understanding May lack depth in General overview
of federated learning. technical aspects.
[10] Expands on advantages such as data May not offer novel Edge devices
privacy and scalability. insights beyond existing
literature.
[4] Offers a detailed overview of Might lack exploration of General overview
collaborative model training in newer advancements.
federated learning.
[18] Addresses challenges in federated May not provide practical General overview
learning, offering insights into solutions.
achieving desired model performance.
[3] Highlights the significance of federated May not delve deeply into General overview
learning in preserving data privacy, technical aspects.
especially in sensitive domains.
[1] Explores efficient utilization of May lack in-depth General overview
distributed computing resources in analysis of other aspects.
federated learning.
[19] Discusses the applications of federated Might not extensively Healthcare
learning across various domains, cover technical
showcasing its versatility. challenges.
[8] Highlights the adaptability of federated May not provide concrete Computer vision
learning to diverse data distributions examples of
and privacy requirements. applications.
[11] Addresses privacy concerns and Might not explore newer Privacy-preserving
scalability in federated learning, advancements. techniques
particularly focusing on distributed
data sources.
[20] Discusses how federated learning May not provide Recommender
accommodates data heterogeneity comprehensive insights systems
while preserving privacy and integrity. into other aspects.
[21] Highlights ongoing research efforts to May not provide practical Security
address challenges such as model implementations.
synchronization and security
vulnerabilities.
[22] Explores the potential of federated May not deeply analyze Wireless
learning in dynamic and resource- technical intricacies. communication
constrained environments like IoT
networks.
[13] Expands the scope of federated learning May not provide in-depth Smart grids
to include advanced techniques like exploration of advanced
federated reinforcement learning. techniques.
[23] Explores the applications of federated Might not extensively Social good
learning in domains like robotics and discuss technical applications
personalized AI assistants. challenges.
Applications of Federated Learning 181

FEDERATED LEARNING TAXONOMY WITH AI

Federated learning, an exponentially growing field at the intersection of machine
learning and distributed systems, encompasses a diverse range of methodologies and
techniques aimed at collaborative model training across decentralized data sources.
To provide a comprehensive understanding of federated learning, it can be catego-
rized into several dimensions, each reflecting different aspects of the training pro-
cess and its applications in various domains.

Federated Learning Architectures

Federated learning architectures constitute the foundational framework upon which
collaborative model training across distributed data sources is built. These architec-
tures are pivotal in determining how data is partitioned and utilized across devices
or servers while ensuring efficient model updates and aggregation. Horizontal fed-
erated learning, for instance, involves partitioning data samples across multiple
devices or servers, allowing each participant to train the model locally on its subset
of data [8]. On the other hand, vertical federated learning partitions feature vertically
across devices, enabling collaborative training on different aspects of the data [10].
Additionally, federated transfer learning extends the scope of federated learning by
facilitating knowledge transfer from one domain to another while preserving data
privacy and security [4].
Alongside this, federated learning architectures continue to evolve to address the
specific needs and challenges of different application domains. For instance, feder-
ated learning in healthcare often adopts vertical federated learning architectures to
enable collaborative model training across disparate healthcare institutions while
ensuring patient data privacy and compliance with regulatory requirements such as
HIPAA [19]. Similarly, in the financial sector, horizontal federated learning archi-
tectures are leveraged for collaborative fraud detection and risk assessment across
multiple banks or financial institutions while safeguarding customer privacy [3].
These diverse architectures underscore the versatility and adaptability of federated
learning to different data distribution scenarios and privacy constraints.
In summary, federated learning architectures play a crucial role in facilitat-
ing collaborative model training across decentralized data sources. By embracing
different partitioning strategies such as horizontal, vertical, and transfer learning,
federated learning architectures empower organizations to harness the collective
intelligence of distributed data while respecting individual privacy rights and regu-
latory constraints.

Federated Learning Algorithms

In the realm of federated learning, various algorithms have been developed to opti-
mize the collaborative model training process across decentralized data sources.
These algorithms play a crucial role in addressing challenges such as communica-
tion overhead, model convergence, and privacy preservation. One of the prominent
algorithms is federated averaging, which involves aggregating model updates from
182 Artificial Intelligence Using Federated Learning

FIGURE 10.1 Federated averaging (FedAvg).

multiple devices or servers to compute a global model while mitigating the impact of
data heterogeneity and communication constraints [5]. Additionally, FedProx inte-
grates proximal terms into the optimization objective to enhance model convergence
and stability in federated learning settings [24]. FedAvg+ extends the capabilities of
federated averaging by incorporating client selection strategies and adaptive learn-
ing rates to improve model performance and convergence speed [25]. Figure 10.1
illustrates the main concept of FedAvg.
Added to that, secure aggregation algorithms have been developed to guarantee
privacy and confidentiality throughout the model aggregation procedure in feder-
ated learning. These algorithms leverage cryptographic methods like secure multi-
party computation (SMC) and homomorphic encryption to securely aggregate model
updates from multiple participants without exposing sensitive information [3]. By
encrypting model updates before transmission and decrypting them only during the
aggregation phase, secure aggregation algorithms provide robust privacy guarantees
while enabling collaborative model training across distributed data sources.
Furthermore, federated learning algorithms continue to evolve to address emerg-
ing challenges and application-specific requirements. For instance, in federated
learning for IoT networks, lightweight and communication-efficient algorithms are
essential to accommodate resource-constrained edge devices and minimize energy
consumption [26]. Similarly, in federated learning for healthcare informatics, algo-
rithms that prioritize patient privacy and compliance with regulatory standards
such as GDPR and HIPAA are paramount [19]. By developing and refining feder-
ated learning algorithms tailored to specific use cases and scenarios, researchers
and practitioners can unlock the full potential of collaborative model training across
decentralized data sources while ensuring privacy, efficiency, and scalability.

Privacy-Preserving Techniques
Maintaining privacy is of utmost importance in federated learning, given the decen-
tralized nature of data sources and the sensitive nature of the information being pro-
cessed. To address this challenge, various privacy-preserving techniques have been
developed and integrated into federated learning frameworks. One such technique is
Applications of Federated Learning 183

differential privacy, which ensures that individual data contributions remain private
by introducing noise to the gradients or model updates before aggregation [5]. By
introducing controlled randomness, differential privacy prevents malicious actors
from inferring sensitive information about individual data samples while still allow-
ing for accurate model training at the aggregate level.
Another privacy-preserving technique widely employed in federated learning is
homomorphic encryption, which allows computations to be conducted on encrypted
data without the need for decryption. By encrypting model parameters and updates
before transmission and performing aggregation operations in the encrypted domain,
homomorphic encryption ensures end-to-end privacy and confidentiality of the data
throughout the federated learning process [3]. Similarly, secure multiparty computa-
tion permits multiple parties to collectively compute a function using their individ-
ual inputs, without disclosing any details about the inputs. themselves. In federated
learning, SMC can be used to securely aggregate model updates from multiple par-
ticipants without disclosing individual data samples or model parameters [27].
Furthermore, federated learning frameworks often employ federated learning
with encrypted gradients (FLEG) to further enhance privacy and security. FLEG
involves encrypting the gradients of the local model updates before transmission and
decrypting them only during the aggregation phase, ensuring that sensitive infor-
mation remains protected throughout the communication process [28]. By combin-
ing encryption techniques with federated learning methodologies, FLEG provides
robust privacy guarantees while enabling collaborative model training across dis-
tributed data sources.
Overall, privacy-preserving techniques play a crucial role in ensuring the confi-
dentiality and integrity of data in federated learning settings. By leveraging methods
such as differential privacy, homomorphic encryption, secure multiparty computa-
tion, and federated learning with encrypted gradients, organizations can harness the
benefits of collaborative model training while safeguarding individual privacy rights
and complying with regulatory requirements.

FEDERATED LEARNING APPLICATION AREAS

Federated learning is rapidly gaining traction across various domains, offering inno-
vative solutions to address data privacy concerns, scalability challenges, and regula-
tory compliance requirements.
Federated learning is revolutionizing the landscape of the Internet of Things by
enabling collaborative model training across decentralized edge devices while preserv-
ing data privacy and security [5]. In IoT applications, where data is generated at the edge
by a multitude of interconnected devices, FL offers a distributed framework for model
training, eliminating the need to transmit sensitive data to centralized servers [3].
One prominent application of federated learning in IoT is in smart energy manage-
ment systems. In this scenario, sensors and smart meters deployed in homes, build-
ings, and industrial facilities collect real-time energy consumption data. Federated
learning enables collaborative model training across these distributed sensors to
develop predictive models for energy demand forecasting, anomaly detection, and
optimization of energy usage [29]. By leveraging FL, energy providers can enhance
184 Artificial Intelligence Using Federated Learning

grid stability, reduce energy waste, and optimize renewable energy integration while
preserving consumer privacy and data confidentiality [30].
Another compelling application of federated learning in IoT is predictive main-
tenance for industrial machinery and equipment. In manufacturing environments,
sensors and actuators embedded in machinery collect operational data, such as tem-
perature, vibration, and pressure. Federated learning allows these edge devices to
collaboratively train machine learning models for predictive maintenance, enabling
early detection of equipment failures, reducing downtime, and minimizing main-
tenance costs [31]. By leveraging FL, manufacturers can improve operational effi-
ciency, extend equipment lifespan, and enhance overall productivity [32].
Federated learning holds promise for environmental monitoring applications in
IoT, where sensors are deployed to collect data on air quality, water quality, and
climate conditions. Federated learning enables collaborative model training across
distributed sensors to develop predictive models for environmental forecasting, pol-
lution detection, and disaster management [33]. By leveraging FL, environmental
agencies can gain insights into complex environmental phenomena, facilitate timely
interventions, and safeguard public health and safety [34].

FIGURE 10.2 Federated learning applications.

Applications of Federated Learning 185

Federated learning presents a transformative approach to model training and

inference in IoT environments, enabling collaborative learning across decentralized
edge devices while addressing privacy, latency, and resource constraints. By har-
nessing the power of FL, organizations can unlock the full potential of IoT data for
applications ranging from energy management and predictive maintenance to envi-
ronmental monitoring and beyond. Figure 10.2 shows some FL application areas that
are worth mentioning.

Federated Learning Applications in Healthcare

Federated learning holds immense promise in revolutionizing healthcare by enabling
collaborative model training on sensitive medical data while preserving patient pri-
vacy and complying with regulatory requirements. One of the key applications of FL
in healthcare is in the development of personalized treatment recommendations and
disease prediction models. By aggregating insights from diverse patient populations
across multiple healthcare institutions, FL facilitates the creation of robust predictive
models that can identify patterns and risk factors associated with various medical
conditions, such as cardiovascular diseases, cancer, and diabetes [35]. These predic-
tive models enable healthcare providers to deliver personalized interventions and
preventive measures tailored to individual patient needs, leading to improved clini-
cal outcomes and patient satisfaction [36].
Another significant application of FL in healthcare is in clinical decision sup-
port systems (CDSSs), where collaborative model training enables the development
of intelligent algorithms that assist healthcare professionals in diagnosing diseases,
interpreting medical images, and recommending treatment options. By leverag-
ing FL, CDSS can integrate insights from heterogeneous data sources, including
electronic health records (EHRs), medical imaging data, genomic data, and wear-
able sensor data, to provide more accurate and timely clinical decision support [37].
Moreover, FL-powered CDSSs can enhance diagnostic accuracy, reduce medical
errors, and improve treatment efficacy, ultimately leading to better patient care and
outcomes [38].
FL plays a crucial role in medical research and drug discovery by enabling collab-
orative analysis of large-scale biomedical datasets while protecting patient privacy
and confidentiality. Pharmaceutical companies and research institutions can lever-
age FL to aggregate insights from distributed clinical trials, genomic studies, and
real-world patient data to accelerate the discovery of novel therapies and drug targets
[39]. Additionally, FL facilitates federated learning meta-analysis, where aggregated
models from multiple sites are combined to derive insights and validate findings
across diverse patient populations, leading to more robust and generalizable research
outcomes [40].
FL contributes to population health management by enabling population-level
analytics and predictive modeling to identify disease trends, allocate healthcare
resources, and implement targeted interventions. Public health agencies and health-
care organizations can leverage FL to analyze aggregated data from diverse sources,
including public health records, environmental data, and social determinants of
health, to identify high-risk populations, detect disease outbreaks, and design
186 Artificial Intelligence Using Federated Learning

effective public health interventions [41]. By harnessing the collective intelligence of

distributed data sources, FL empowers stakeholders to make informed decisions and
take proactive measures to improve population health and well-being.
Federated learning applications in healthcare encompass a wide range of use
cases, including personalized medicine, clinical decision support, medical research,
and population health management. By enabling collaborative model training on
decentralized data sources while ensuring privacy and compliance, FL facilitates
advancements in healthcare delivery, medical research, and public health initiatives,
ultimately leading to better outcomes for patients and communities.

Federated Learning Applications in Cybersecurity

Federated learning emerges as a promising approach to strengthen cybersecurity
defenses by harnessing the collective intelligence of distributed devices and networks
while preserving data privacy and confidentiality. In the cybersecurity domain, FL
facilitates collaborative threat detection, anomaly detection, and malware analysis
across diverse endpoints, networks, and security platforms. By aggregating insights
from multiple sources without sharing raw data, FL enables organizations to detect
and mitigate cyber threats in real time while minimizing the risk of data breaches
and privacy violations [42].
One of the key applications of FL in cybersecurity is collaborative threat intel-
ligence sharing, where security information and event management (SIEM) sys-
tems across different organizations collaborate to identify and respond to emerging
threats. FL enables SIEM systems to collectively analyze security logs, network traf-
fic, and endpoint telemetry data to detect malicious activities, unauthorized access
attempts, and insider threats [43]. By leveraging FL techniques, organizations can
enhance their threat detection capabilities, reduce false positives, and respond to
cyber incidents more effectively, thereby strengthening their overall cybersecurity
posture.
FL facilitates federated anomaly detection by enabling distributed sensors,
devices, and security platforms to collaboratively identify abnormal patterns and
behaviors indicative of potential security breaches or cyber-attacks. By aggregating
local anomaly detection models trained on diverse data sources, FL enables organi-
zations to detect sophisticated threats such as zero-day exploits, advanced persistent
threats (APTs), and insider attacks [44]. Additionally, FL enables adaptive threat
modeling by continuously updating and refining anomaly detection models based on
evolving cyber threats and attack techniques, thereby improving detection accuracy
and resilience against emerging threats.
Federated learning plays an important role in federated malware analysis by
enabling security researchers and analysts to collaboratively analyze and classify
malware samples while preserving the confidentiality of sensitive information. By
sharing insights from malware analysis models trained on distributed data sources,
FL enables organizations to identify new malware variants, detect poly-morphic
malware, and generate actionable threat intelligence to enhance their defense strat-
egies [45]. Additionally, FL facilitates federated intrusion detection by enabling
distributed intrusion detection systems (IDSs) to collaboratively analyze network
Applications of Federated Learning 187

traffic, identify suspicious activities, and block malicious traffic in real time, thereby
safeguarding networks from cyber threats and unauthorized access attempts [46].
Federated learning applications in cybersecurity offer innovative solutions
to address the evolving threat landscape and protect organizations against cyber-
attacks, data breaches, and privacy violations. By leveraging collaborative threat
detection, anomaly detection, malware analysis, and intrusion detection techniques,
FL enables organizations to enhance their cybersecurity defenses, mitigate risks, and
ensure the integrity and confidentiality of their sensitive information.

Federated Learning Applications in Smart Homes and Smart Cities

Federated learning is revolutionizing the development of smart homes and smart
cities by enabling collaborative intelligence while safeguarding user privacy and
data security. In smart homes, FL facilitates personalized automation and energy
management by leveraging insights from distributed devices and sensors without
compromising individual privacy. By aggregating local data from smart appliances,
thermostats, and energy meters, FL enables homeowners to optimize energy con-
sumption, reduce utility costs, and enhance comfort levels while preserving sensitive
information within the confines of their homes [47].
It enables collaborative predictive maintenance in smart homes by leveraging
insights from connected devices to anticipate equipment failures, identify mainte-
nance needs, and optimize maintenance schedules. By pooling together data from
smart HVAC systems, home security cameras, and IoT devices, FL enables proactive
maintenance strategies that minimize downtime, reduce repair costs, and prolong
the lifespan of home appliances and infrastructure [48].
In the context of smart cities, FL plays a pivotal role in optimizing urban infra-
structure, enhancing public safety, and improving the quality of life for residents. FL
enables collaborative traffic management by aggregating insights from distributed
sensors, cameras, and traffic signals to optimize traffic flow, reduce congestion, and
mitigate traffic accidents in real time. By analyzing traffic patterns, pedestrian flows,
and environmental factors, FL empowers city planners to implement data-driven inter-
ventions that improve transportation efficiency and reduce environmental impact [49].
Moreover, FL facilitates smart grid optimization by enabling collaborative
energy management across distributed power generation, distribution, and consump-
tion nodes. By aggregating insights from smart meters, renewable energy sources,
and energy storage systems, FL enables utilities to balance supply and demand, opti-
mize grid operations, and integrate renewable energy resources more effectively.
Additionally, FL enables demand response programs that incentivize consumers to
adjust their energy usage patterns in response to grid conditions, thereby promoting
energy conservation and grid stability [50].
Federated learning applications in smart homes and smart cities offer transforma-
tive solutions to address energy efficiency, infrastructure optimization, and urban
sustainability challenges. By leveraging collaborative intelligence while preserving
data privacy and security, FL enables homeowners, city planners, and utility provid-
ers to unlock the full potential of connected devices and IoT technologies in creating
smarter, more sustainable living environments.
188 Artificial Intelligence Using Federated Learning

Federated Learning Applications in Industry 4.0

Federated learning is poised to revolutionize the landscape of Industry 4.0 by
enabling collaborative intelligence among interconnected machines, robots, and
industrial processes while preserving data privacy and security. In the context of
smart manufacturing, FL facilitates predictive maintenance by leveraging insights
from distributed sensors, IoT devices, and equipment without exposing sensitive
operational data to external parties. By aggregating local data from manufacturing
machines, FL enables predictive maintenance models to identify potential equip-
ment failures, schedule maintenance activities, and optimize production schedules,
thereby reducing downtime and minimizing production disruptions [51].
FL enables collaborative quality control in smart factories by aggregating insights
from distributed inspection systems, cameras, and sensors to detect defects, anoma-
lies, and deviations in real time. By analyzing images, sensor readings, and process
parameters, FL empowers manufacturers to identify quality issues early in the pro-
duction process, adjust production parameters dynamically, and prevent defective
products from reaching the market, thereby improving product quality and customer
satisfaction [52].
FL facilitates federated robotics by enabling collaborative model training across
distributed robots and autonomous systems while preserving the confidentiality of
proprietary algorithms and operational data. By aggregating insights from robot tra-
jectories, sensor data, and environmental conditions, FL enables robots to learn from
each other’s experiences, adapt to changing production environments, and optimize
task execution strategies collaboratively [53].
Moreover, FL plays a crucial role in supply chain optimization by enabling col-
laborative demand forecasting, inventory management, and logistics planning across
distributed nodes in the supply chain network. By aggregating insights from retailers,
suppliers, and logistics partners, FL enables supply chain stakeholders to improve
demand forecasting accuracy, reduce inventory carrying costs, and optimize trans-
portation routes and delivery schedules, thereby enhancing supply chain efficiency
and responsiveness [54].
In summary, federated learning applications in Industry 4.0 offer transforma-
tive solutions to address operational challenges, enhance productivity, and optimize
resource utilization across various industrial sectors. By leveraging collaborative
intelligence while preserving data privacy and security, FL enables manufacturers,
robotics companies, and supply chain stakeholders to harness the power of distrib-
uted data and interconnected systems in creating smarter, more efficient industrial
ecosystems.

Federated Learning Applications in Autonomous Driving

Federated learning is poised to transform the landscape of autonomous driving by
enabling collaborative model training across distributed vehicles while ensuring
data privacy and security. In the context of autonomous vehicles (AVs), FL facili-
tates collaborative perception by aggregating insights from onboard sensors, cam-
eras, and LiDAR systems to enhance object detection, lane detection, and semantic
Applications of Federated Learning 189

segmentation capabilities. By leveraging local data from diverse driving scenarios

and environmental conditions, FL enables AVs to learn from each other’s expe-
riences, improve perception accuracy, and adapt to complex real-world driving
conditions [55].
Alongside this, FL enables federated localization by leveraging insights from GPS
data, inertial measurement units (IMUs), and map information to enhance localiza-
tion accuracy and reliability in GPS-denied environments such as urban canyons
and tunnels. By aggregating localization models trained on distributed vehicles, FL
enables AVs to localize themselves relative to the surrounding environment more
accurately, navigate complex road networks, and maintain lane-level precision even
in challenging scenarios [56].
FL plays a significant role in federated planning and decision-making by enabling
collaborative model training across AVs to optimize trajectory planning, lane chang-
ing, and merging maneuvers in real-time. By aggregating insights from diverse driv-
ing behaviors and traffic conditions, FL enables AVs to anticipate potential hazards,
avoid collisions, and navigate safely in complex traffic environments while adhering
to traffic rules and regulations [57].
Additionally, FL facilitates federated reinforcement learning by enabling col-
laborative model training across AVs to improve adaptive driving behaviors and
decision-making strategies. By aggregating insights from diverse driving experiences
and scenarios, FL enables AVs to learn from both positive and negative outcomes,
refine driving policies, and adapt to changing road conditions, traffic patterns, and
user preferences over time [58].
Federated learning applications in autonomous driving offer transformative solu-
tions to enhance perception, localization, planning, and decision-making capabilities
of AVs while ensuring data privacy, security, and regulatory compliance. By leverag-
ing collaborative intelligence across distributed vehicles, FL enables AVs to navigate
safely and efficiently in diverse driving environments, paving the way for the wide-
spread adoption of autonomous transportation systems in the future.

Federated Learning Applications in Metaverse and Virtual Reality

Federated learning is poised to revolutionize the development of the metaverse and
virtual reality (VR) environments by enabling collaborative model training across
distributed users while preserving data privacy and security. In the context of the
metaverse, FL facilitates collaborative content creation by aggregating insights from
distributed users’ interactions, preferences, and behaviors to generate personalized
virtual experiences. By leveraging local data from diverse user interactions, FL
enables metaverse platforms to customize content recommendations, optimize user
interfaces, and enhance immersion levels based on individual preferences and usage
patterns [59].
FL enables federated scene understanding by leveraging insights from distributed
users’ viewpoints, movements, and interactions to improve 3D scene reconstruction,
object detection, and semantic segmentation in VR environments. By aggregating
scene understanding models trained on diverse user perspectives, FL enables VR
applications to create more realistic and interactive virtual worlds, enhance spatial
190 Artificial Intelligence Using Federated Learning

awareness, and enable collaborative experiences among users with different devices
and interaction modalities [60].
FL plays a crucial role in federated avatar customization by enabling collabora-
tive model training across distributed users to generate personalized avatars based
on individual preferences, physical attributes, and style preferences. By aggregating
avatar customization models trained on diverse user datasets, FL enables metaverse
platforms to create more diverse and inclusive virtual communities, enhance user
representation, and foster social interactions and engagement in virtual environ-
ments [61].
Additionally, FL facilitates federated emotion recognition by leveraging insights
from distributed users’ facial expressions, gestures, and physiological signals to infer
emotional states and reactions in VR experiences. By aggregating emotion recogni-
tion models trained on diverse user interactions, FL enables VR applications to adapt
content and interactions dynamically based on users’ emotional responses, enhance
storytelling, and create more immersive and engaging virtual experiences [62].
Federated learning applications in the metaverse and virtual reality offer transfor-
mative solutions to enhance content creation, scene understanding, avatar custom-
ization, and emotion recognition capabilities while ensuring data privacy, security,
and user autonomy. By leveraging collaborative intelligence across distributed users,
FL enables metaverse platforms to create more personalized, immersive, and inter-
active virtual environments, shaping the future of digital entertainment, social inter-
action, and online collaboration.

PROSPECTS FOR FEDERATED LEARNING AND CHALLENGES

Federated learning holds immense promise for revolutionizing various domains,
offering a decentralized approach to machine learning that preserves data privacy,
enhances collaboration, and enables more efficient model training across distributed
devices and networks. One of the key prospects for FL lies in its potential to democ-
ratize AI by enabling broader participation and knowledge sharing among diverse
stakeholders. By allowing organizations and individuals to collaborate on model
training without sharing raw data, FL empowers smaller players, researchers, and
developers to access and contribute to state-of-the-art AI models and applications,
fostering innovation and inclusivity in the AI ecosystem [63].
FL offers a viable solution to address data privacy concerns and regulatory compli-
ance requirements in increasingly stringent regulatory environments. By keeping sensi-
tive data localized and performing model training on-device or at the edge, FL minimizes
the risks associated with data breaches, unauthorized access, and privacy violations. This
makes FL particularly well-suited for industries such as healthcare, finance, and govern-
ment, where data security and regulatory compliance are paramount [64].
Moreover, FL has the potential to improve the scalability and efficiency of AI
model training by distributing computation and storage requirements across multiple
devices and networks. By lever- aging the computational resources available at the
edge and aggregating insights from diverse data sources, FL enables organizations
to train complex models on large-scale datasets without the need for centralized data
processing or expensive infrastructure. This scalability enables FL to address the
Applications of Federated Learning 191

challenges of training models on increasingly large and heterogeneous datasets, pav-

ing the way for advancements in AI research and applications [65].
Despite its promising prospects, federated learning also faces several challenges
and limitations that need to be addressed to realize its full potential as shown in
Figure 10.3.
One of the primary challenges is the heterogeneity and distribution of data across
different devices and networks, which can lead to issues such as data imbalance,
non-IID (non-independent and identically distributed) data distributions, and model
performance disparities. Addressing these challenges requires developing robust
federated learning algorithms, techniques for data preprocessing, and strategies for
model aggregation and adaptation to accommodate diverse data sources and char-
acteristics [66].
Additionally, ensuring the security and integrity of federated learning systems
remains a significant challenge, particularly in the face of adversarial attacks, model

FIGURE 10.3 Federated learning challenges.

192 Artificial Intelligence Using Federated Learning

poisoning, and data manipulation attempts. Protecting FL systems against privacy

breaches, information leakage, and model vulnerabilities requires implementing
robust security protocols, encryption techniques, and anomaly detection mecha-
nisms to detect and mitigate potential threats and vulnerabilities [67].
Federated learning introduces communication overhead and bandwidth con-
straints, especially in scenarios with large numbers of participating devices or net-
works. Optimizing communication protocols, compression techniques, and model
update strategies are essential to minimize communication costs, reduce latency,
and improve the efficiency of federated learning systems, particularly in resource-
constrained environments such as mobile devices and IoT networks [68].
In conclusion, federated learning offers promising prospects for democratizing
AI, preserving data privacy, and improving the scalability and efficiency of model
training across distributed devices and networks. However, addressing the chal-
lenges of data heterogeneity, security, and communication overhead is crucial to
realizing the full potential of federated learning and unlocking its transformative
impact across various domains.

CONCLUSION
This chapter shows how far federated learning represents a groundbreaking para-
digm shift in machine learning, offering a decentralized approach to model train-
ing that preserves data privacy, enhances collaboration, and enables more efficient
utilization of distributed resources. Throughout various domains, from healthcare to
finance, from smart cities to autonomous vehicles, FL has demonstrated its poten-
tial to address the challenges of data privacy, scalability, and regulatory compliance
while unlocking new opportunities for innovation and collaboration.
As we look to the future, the prospects for federated learning are bright. With
advancements in federated learning algorithms, communication protocols, and secu-
rity mechanisms, FL is poised to become a cornerstone of the AI landscape, democ-
ratizing access to AI technologies and empowering organizations and individuals to
leverage collective intelligence for solving complex problems.
However, federated learning also presents challenges that must be addressed to
realize its full potential. Issues such as data heterogeneity, security vulnerabilities,
and communication overhead require ongoing research and development efforts to
develop robust solutions that can ensure the reliability, scalability, and security of
federated learning systems.
In conclusion, federated learning holds tremendous promise for shaping the future
of AI, enabling collaborative intelligence across distributed devices and networks
while preserving data privacy and security. By fostering collaboration, innovation,
and inclusivity, federated learning has the potential to drive transformative advance-
ments in AI research, applications, and industries, paving the way for a more con-
nected, intelligent, and equitable future.

REFERENCES
[1] Jakub Konecný et al. Federated learning: Strategies for improving communication effi-
ciency. ArXiv abs/1610.05492, 2016.
Applications of Federated Learning 193

[2] Xinyi Li et al. Federated learning for healthcare informatics. Journal of Healthcare
Informatics Research, 5(1):1–23, 2021.
[3] Keith Bonawitz et al. Towards federated learning at scale: System design. arXiv pre-
print arXiv:1902.01046, 2019.
[4] Peter Kairouz et al. Advances and open problems in federated learning. arXiv preprint
arXiv:1912.04977, 2019.
[5] H. Brendan McMahan et al. Federated learning: Collaborative machine learning with-
out centralized training data. ACM Transactions on Machine Learning and Data
Mining (TOMM), 10(2):1–21, 2019.
[6] Qiang Yang et al. Federated Learning. Synthesis Lectures on Artificial Intelligence and
Machine Learning, 13(3):1–207, 2020.
[7] Yang Liu et al. Federated learning for natural language processing. arXiv preprint
arXiv:1910.13067, 2019.
[8] Qiang Yang et al. Federated learning for computer vision: A survey. arXiv preprint
arXiv:2008.08102, 2020.
[9] Tianqing Li et al. Federated learning for personalized recommendation: A review.
Information Fusion, 73:22–38, 2022.
[10] Tian Li et al. Federated learning in edge computing: A comprehensive survey. IEEE
Communications Surveys & Tutorials, 22(3):2031–2063, 2020.
[11] Tongxin Zhao et al. Privacy-preserving federated learning: A survey. IEEE Internet of
Things Journal, 8(7):5217–5242, 2021.
[12] R. Zhang Jingxin Mao, Hanqiu Wang, Bing Li, Xiang Cheng, Liuqing Yang. A sur-
vey on federated learning in intelligent transportation systems. IEEE Transactions on
Intelligent Vehicles, 1–17, 2024.
[13] Jian Li and Tongbao Chen and Shaohua Teng. A comprehensive survey on client selec-
tion strategies in federated learning. Computer Networks, 251:110663, 2021.
[14] Yuheng Xie et al. Federated learning for pharmaceutical research: Opportunities and
challenges. arXiv preprint arXiv:2105.10644, 2021.
[15] Ingrid Gundersen et al. Federated learning in biotechnology: Advancements and appli-
cations. Frontiers in Genetics, 12:1110, 2021.
[16] Gabriela Nascimento et al. Federated learning for environmental monitoring and con-
servation. IEEE Access, 9:160535–160550, 2021.
[17] S.R. Pokhrel. Federated learning meets blockchain at 6G edge: A drone-assisted net-
working for disaster response. In Proceedings of the 2nd ACM MobiCom Workshop
on Drone Assisted Wireless Communications for 5G and Beyond, pp. 49–54, 2020,
September.
[18] Qiang Yang et al. Federated learning: Challenges, methods, and future directions.
IEEE Transactions on Wireless Communications, 30(1):405–421, 2021.
[19] Xinyi Li et al. Federated learning for healthcare informatics. Journal of Healthcare
Informatics Research, 5(1):1–23, 2021.
[20] Qiang Yang et al. Federated learning for personalized recommendation: A review.
Information Fusion, 73:22–38, 2022.
[21] Cong Li et al. Secure and privacy-preserving federated learning for 5g and beyond:
Challenges, solutions, and future directions. IEEE Wireless Communications,
28(3):132–139, 2021.
[22] Fangming Hu et al. Federated learning in wireless communication networks: Challenges,
methods, and future directions. IEEE Communications Magazine, 59(7):38–45, 2021.
[23] T. Zhang et al. Avestimehr. Federated learning for the internet of things: Applications,
challenges, and opportunities. IEEE Internet of Things Magazine, 5(1):24–29, 2022.
[24] Yazeed Yasin Ghadi et al. Integration of federated learning with IoT for smart cities
applications, challenges, and solutions. PeerJ Computer Science, 9:e1657, 2023.
194 Artificial Intelligence Using Federated Learning

[25] Phoebe Clark et al. Federated AI, current state, and future potential. Asia-Pacific
Journal of Ophthalmology, 12(3):310–314, 2023.
[26] Virginia Smith et al. Fedprox: Federated optimization with proximal term. Advances in
Neural Information Processing Systems, 32:3481–3492, 2019.
[27] Yu-Xiang Wang et al. Fedavg+: Communication-efficient distributed deep learning.
Advances in Neural Information Processing Systems, 33:2495–2506, 2020.
[28] H.B. McMahan et al. Federated learning: Strategies for improving communication
efficiency. In Proceedings of the 29th Conference on Neural Information Processing
Systems (NIPS), Barcelona, Spain, pp. 5–10, 2016, December.
[29] Stacey Truex et al. A hybrid secure multiparty computation and blockchain proto-
col for privacy-preserving federated learning in healthcare. Journal of Biomedical
Informatics, 118:103802, 2021.
[30] Jakub Konečný et al. Federated learning with encrypted gradient aggregation in the
wild. arXiv preprint arXiv:2007.10749, 2020.
[31] A. Imteaj et al. A survey on federated learning for resource-constrained IoT devices.
IEEE Internet of Things Journal, 9(1):1–24, 2021.
[32] X. Yin et al. A comprehensive survey of privacy-preserving federated learning: A tax-
onomy, review, and future directions. ACM Computing Surveys (CSUR), 54(6):1–36,
2021.
[33] Jakub Konečný et al. Federated learning for predictive maintenance in industry 4.0.
IEEE Transactions on Industrial Informatics, 17(2):1282–1290, 2021.
[34] Rafael Abreu et al. Federated learning for predictive maintenance in industry 4.0. IEEE
Transactions on Industrial Informatics, 17(2):1282–1290, 2021.
[35] D.C. Nguyen et al. Federated learning for internet of things: A comprehensive survey.
IEEE Communications Surveys & Tutorials, 23(3):1622–1658, 2021.
[36] J.C. Jiang et al. Federated learning in smart city sensing: Challenges and opportunities.
Sensors, 20(21):6230, 2020..
[37] R.S. Antunes et al. Federated learning for healthcare: Systematic review and archi-
tecture proposal. ACM Transactions on Intelligent Systems and Technology (TIST),
13(4):1–23, 2022.
[38] P. Dhade and P. Shirke. Federated learning for healthcare: A comprehensive review.
Engineering Proceedings, 59(1):230, 2024.
[39] S. Sharma and K. Guleria. A comprehensive review on federated learning based mod-
els for healthcare applications. Artificial Intelligence in Medicine, 146:102691, 2023.
[40] S. Banabilah et al. Federated learning review: Fundamentals, enabling technologies,
and future applications. Information Processing & Management, 59(6):103061, 2022.
[41] A. Rauniyar et al. Federated learning for medical applications: A taxonomy, current
trends, challenges, and future research directions. IEEE Internet of Things Journal,
11(5):7374–7398, 2023.
[42] L. Li et al. A review of applications in federated learning. Computers & Industrial
Engineering, 149:106854, 2020.
[43] D.C. Nguyen et al. Federated learning for smart healthcare: A survey. ACM Computing
Surveys (Csur), 55(3):1–37, 2022
[44] Bo Xu et al. Federated learning for cybersecurity: A comprehensive survey. ACM
Computing Surveys, 54(1):1–36, 2021.
[45] Xun Chen et al. Collaborative threat detection with federated learning in edge comput-
ing. IEEE Internet of Things Journal, 8(22):16184–16196, 2021.
[46] Xiaoyong Wang et al. Federated learning for anomaly detection: A comprehensive
review. Computers & Security, 107:102291, 2021.
[47] Peng Liu et al. Federated learning for malware detection: A survey. Future Generation
Computer Systems, 128:135–146, 2021.
Applications of Federated Learning 195

[48] Xiaoyong Wang et al. Federated learning for intrusion detection: Challenges, methods,
and future directions. Journal of Network and Computer Applications, 197:105028,
2022.
[49] Tong Zhao et al. Federated learning for smart home energy management: A compre-
hensive review. Sustainable Cities and Society, 80:102828, 2022.
[50] S. Pandya et al. Federated learning for smart cities: A comprehensive survey.
Sustainable Energy Technologies and Assessments, 55:102987, 2023.
[51] Z. Zheng et al. Applications of federated learning in smart cities: Recent advances,
taxonomy, and open challenges. Connection Science, 34(1):1–28, 2022.
[52] Jie Li et al. Federated learning for smart grid optimization: Challenges and opportuni-
ties. Energies, 14(5):1330, 2021.
[53] I. Ullah et al. Multi-level federated learning for industry 4.0—A crowdsourcing
approach. Procedia Computer Science, 217:423–435, 2023.
[54] S. Savazzi et al. Opportunities of federated learning in connected, cooperative, and
automated industrial systems. IEEE Communications Magazine, 59(2):16–21, 2021.
[55] Y. Xianjia et al. Federated learning in robotic and autonomous systems. Procedia
Computer Science, 191:135–142, 2021.
[56] J. Zhu et al. Blockchain-empowered federated learning: Challenges, solutions, and
future directions. ACM Computing Surveys, 55(11):1–31, 2023.
[57] V.P. Chellapandi and Liangqi Yuan and Stanisław H. Żak and Ziran Wang. Federated
learning for connected and automated vehicles: A survey of existing approaches and
challenges. IEEE Transactions on Intelligent Vehicles, pp. 2485–2492, 2023.
[58] B. Yuksek et al. Federated meta learning for visual navigation in GPS-denied urban
airspace. In 2023 IEEE/AIAA 42nd Digital Avionics Systems Conference (DASC) (pp.
1–7). IEEE, 2023, October.
[59] V.P. Chellapandi et al. A survey of federated learning for connected and automated
vehicles. In 2023 IEEE 26th International Conference on Intelligent Transportation
Systems (ITSC) (pp. 2485–2492). IEEE, 2023, September.
[60] L. Lei et al. Deep reinforcement learning for autonomous internet of things: Model,
applications and challenges. IEEE Communications Surveys & Tutorials, 22(3):1722–
1760, 2020.
[61] Y. Chen et al. Federated learning for metaverse: A survey. In Companion Proceedings of
the ACM Web Conference 2023 (pp. 1151–1160), Association for Computing Machinery,
2023, April.
[62] C. Zhang et al. A survey on federated learning. Knowledge-Based Systems, 216:106775,
2021.
[63] Y. Wang et al. Social metaverse: Challenges and solutions. IEEE Internet of Things
Magazine, 6(3):144–150, 2023.
[64] S. Pal et al. Development and progress in sensors and technologies for human emotion
recognition. Sensors, 21(16):5554, 2021.
[65] K. Hu et al. Federated learning: A distributed shared machine learning method.
Complexity, 2021(1):8261663, 2021.
[66] T. Li et al. Federated learning: Challenges, methods, and future directions. IEEE Signal
Processing Magazine, 37(3):50–60, 2020.
[67] Keith Bonawitz et al. Towards federated learning at scale: System design. arXiv pre-
print arXiv:1902.01046, 2019.
[68] K. Bonawitz. Towards federated learning at scale: System design. arXiv preprint
arXiv:1902.01046, 2019.
 11 Exploring Future
Trends and Emerging
Applications
A Glimpse Into
Tomorrow’s Landscape
Utpal Ghosh and Shrabanti Kundu

INTRODUCTION
In this rapidly evolving technological landscape, society finds itself in an era of
unprecedented change and innovation. Each day, artificial intelligence permeates
deeper into human lives, blurring the boundaries between human and machine,
while a host of emerging technologies reshape the very fabric of our reality. From
the omnipresent Internet of Things (IoT) to the cutting-edge realm of neuromor-
phic hardware, technology’s impact is profound and transformative. While numer-
ous new and innovative technologies are currently in development, some stand out
for their practicality and promise in this burgeoning landscape. According to the
empirically validated insights of Gartner, a select group of technologies exhibits the
most potential to yield significant advantages in the coming years. In the next 5 to
10 years, particular technologies such as artificial intelligence (AI), transparently
immersive experiences, and digital platforms are anticipated to seize the spotlight
in the technological landscape. Gartner has pinpointed crucial technologies includ-
ing IoT, serverless platform as a service (PaaS), quantum computing, and software-
defined security [1]. Edge computing marks its entry into the Gartner hype cycle,
presenting approaches to enhance cloud service efficiency through the utilization
of intelligent computational techniques, with a particular focus on locations such as
mobile users or embedded microsystems. Looking forward, the next decade is pre-
dicted to witness “artificial intelligence everywhere” emerge as the most disruptive
technology, fueled by unprecedented computing power, adaptive neural networks,
and the vast reservoirs of big data [2]. As the exploration unfolds, the convergence
of diverse technologies becomes a focal point, transcending the silos of individual
advancements. The emergence of edge computing takes center stage, making its
inaugural appearance in the Gartner hype cycle. This paradigm shift promises to
revolutionize cloud service performance, strategically targeting locations such as
mobile users and embedded microsystems. The interconnected fabric of technology,

196 DOI: 10.1201/9781003482000-11

Exploring Future Trends and Emerging Applications 197

encompassing commercial drones, 5G adoption, human augmentation, and quantum

computing, showcases a tapestry woven with the threads of innovation. The integra-
tion of the Internet of Things with edge computing amplifies the potential for smart
homes, cities, healthcare, agriculture, and transportation, reshaping the very fabric
of these domains. This chapter navigates the intricate landscape of these technolo-
gies, unraveling the synergies and potential that lie within.
Moreover, the narrative extends into the realms of artificial general intelligence
(AGI), deep learning, and machine learning, depicting a future where intelligent sys-
tems pervade every facet of our existence. The discussion on virtual reality and its
role in situated learning adds another layer, emphasizing the interactive experiences
that technology affords. The overarching theme emphasizes not only the promise of
technological advancements but also the imperative for active participation, adapta-
tion, and responsible harnessing of these innovations. The journey into the future,
as portrayed in this exploration, beckons readers to embrace the unfolding techno-
logical narrative with a blend of anticipation, curiosity, and proactive engagement.
Numerous of these advancements leverage the capabilities of the Internet of Things,
while others amalgamate multiple technologies to exploit synergies. This chapter
succinctly outlines such revolutionary technologies that hold the potential to contrib-
ute to the creation of a smarter world. The subsequent sections of this chapter are
structured as follows: the second section provides an elucidation of the Gartner hype
cycle, enumerating technologies teetering on the brink and those at the pinnacle.
Furthermore, examines the time it takes for a technology to evolve from its initial
concept to widespread use, particularly highlighting those technologies that generate
excessive excitement. The third section delves into emerging trends and technolo-
gies, while the fourth section concludes with key findings.

LITERATURE REVIEW
Robotic technology is widely utilized across diverse industries such as manufactur-
ing, healthcare, agriculture, transport, and logistics [3–6]. In manufacturing, robots
excel in tasks like assembly, packaging, and welding due to their precision, speed,
and consistency. Healthcare professionals utilize robots for diagnostics, surgeries,
and rehabilitation, while agricultural robots enhance efficiency and cost-effective-
ness in planting, harvesting, and fertilizing crops. In transport and logistics, robots
streamline operations in automated warehouses and distribution centers ensuring
faster and more accurate order processing. Improved technology for sensors [7],
artificial intelligence, and machine learning which transform robotic capabilities
beyond conventional automated processes are important technologies propelling
robotic breakthroughs. Robots can sense and communicate with the surroundings
using sensors, while AI and ML give them better decision-making capabilities. To
achieve higher degrees of automation and flexibility in robots, these technologies
are essential. To satisfy changing industrial needs and increase efficiency, robots
must be able to analyze large datasets, learn from mistakes, and modify their activi-
ties accordingly. This is made possible by AI and ML algorithms [8]. The subject
of smart robotics is rapidly expanding, combining artificial intelligence, machine
learning, and sophisticated technologies for sensing to allow robots to observe,
198 Artificial Intelligence Using Federated Learning

communicate, gain knowledge, and evolve on their own [9]. Its goal is to create
robots that can seamlessly collaborate with humans and possess sophisticated cogni-
tive capacities. Robotics has the potential to revolutionize companies by increasing
consumer satisfaction, productivity, and effectiveness as technology develops [10].
Nonetheless, attention must be given to dealing with ethical issues, data privacy
issues, economic expulsion, and environmental effects [11–14].
To understand the present status and future directions of intelligent robots, this
comprehensive review of the literature attempts to thoroughly analyze emerging
technologies and developments in the field. It achieves this by addressing signifi-
cant research issues. Through a comprehensive analysis of extant literature, the
writers want to pinpoint obstacles, prospects, and optimal methodologies for creat-
ing and executing robotic approaches in various industries. The potential effects of
intelligent robotics on various sectors as well as society serve as the driving force
behind this research, emphasizing the significance of keeping up with technological
developments and the resulting consequences for responsible integration strategies,
ethical considerations, and effective risk management [15–20]. Enhancing organi-
zational and individual outcomes has been demonstrated through the integration
of emerging technologies powered by data analytics and machine learning algo-
rithms, such as artificial intelligence, blockchain, virtual reality, robotics, Internet
of Things, and quantum computing [21]. The management literature recognizes
that these technologies have significantly improved business performance [22, 23]
and calls on academics to use them to construct research theories. By identify-
ing causal links, data science is being incorporated into management research to
help researchers give more comprehensive solutions to long-standing problems [24].
Even though these technologies provide a wealth of study options, organizational
behavior (OB) has only lately started to delve into this area.
The difficulties in performing open-ended research on new technologies can be
attributed to a lack of experience with machine learning and data science methods
[25], in addition to the technical aspects of computing and computational model-
ing. Present-day OB research initiatives are often very logical and theory-based.
Addressing deficiencies and ambiguities in a hypothesis or phenomenon has a greater
impact on a theoretical-focused methodology than does taking human experiences
into account [26]. Finding a balance across concepts and information integration may
help reveal the deep intricacies of employee behaviors, especially with the wealth of
data and sophisticated computational capabilities accessible today. It could be helpful
to review the present theory-driven agendas in behavioral research as new technology
phenomena in the workplace impact employee attitudes and behaviors. Academics
push for updated research initiatives that seek to advance theory in the analysis of
technology advancements in businesses [27], such as using abductive reasoning to
look at AI-based decision-making [28]. Open innovation, machine learning, AI,
robots, the Internet of Things, virtual reality (VR), and quantum computing [29] are
examples of modern management phenomena that stand to benefit from the suggested
decoupling that makes it possible for new paradigms to be introduced. It is necessary
to relax the current interdependencies that underpin traditional phenomenon-based
theorizing to promote technology-oriented research in OB. To create new paradigms
for the subject, academics have previously promoted doing a nontraditional study of
Exploring Future Trends and Emerging Applications 199

management in a variety of situations concentrating on phenomena outside of man-

agement [30]. The article [31] addresses organizational neuroscience as a novel para-
digm for investigation and provides instructions for incorporating OB ideas into the
“black box” of the brain. The goal of micro-organizational academics is to improve
scientific knowledge by continuously advocating for the combined application of
micro- and macro-level hypotheses, frameworks, and strategies as well as for closing
the gap between behavioral management research and practice [32, 33].

GARTNER HYPE CYCLE

A hype cycle provides a thorough depiction of how specific technologies evolve,
illustrating their emergence, adoption, maturation, and impact on various applica-
tions. It is a valuable perspective for chief information officers (CIOs) and senior IT
leaders as they navigate the transition to digital businesses. Figure 11.1 illustrates
the Gartner hype cycle for emerging technologies in the year 2023, showcasing the
preeminent technologies propelled by industry enthusiasm and evolving trends. The
Gartner hype cycle for emerging technologies in 2023 offers valuable insights into
the maturity and potential impact of various cutting-edge technologies. This annual
report serves as a guide for technology stakeholders, providing a roadmap to navi-
gate the complexities of emerging trends.

FIGURE 11.1 Gartner hype cycle for emerging technologies, 2023.

Source: www.gartner.com/en/articles/what-s-new-in-the-2023-gartner-hype-cycle-for-emerging-
technologies.
200 Artificial Intelligence Using Federated Learning

A summary of some key highlights from the 2023 edition is as follows:

2.1 Artificial Intelligence: AI remains a central focus, as progress in machine
learning, natural language processing, and computer vision propels inno-
vation across various industries. While certain AI applications have
reached maturity, others are still in the early stages of development.
2.2 Blockchain: Blockchain technology remains on the radar, with an ongoing
exploration of its potential in areas such as decentralized finance (DeFi),
supply chain management, digital identity, and smart contracts. Despite
the initial hype, practical implementations are gradually gaining traction.
2.3 Quantum Computing: Quantum computing garners significant atten-
tion for its transformative potential in solving complex problems that
are beyond the capabilities of classical computers. While still nascent,
quantum computing is poised to revolutionize fields such as cryptology,
optimization, and pharmaceutical research.
2.4 Cybersecurity: With the proliferation of cyber threats, cybersecurity
technologies like zero-trust architecture, software-defined security, and
decentralized identity solutions are gaining prominence. Organizations
are increasingly prioritizing robust cybersecurity measures to protect
their digital assets.
2.5 Internet of Things: IoT technologies continue to evolve, with advancements
in edge computing, industrial IoT, and connected devices reshaping indus-
tries such as manufacturing, healthcare, and smart cities. The integration
of IoT devices with AI and analytics drives innovation and efficiency.
2.6 5G Networks: The rollout of 5G networks accelerates the adoption of low-
latency connectivity, and speed, enabling advanced applications such as
virtual reality, autonomous vehicles, and augmented reality (AR). As 5G
infrastructure matures, several new opportunities for innovation emerge.
2.7 Edge Computing: Edge computing architectures gain traction for their
ability to process data closer to the source, reducing latency and band-
width usage. This technology enables real-time analytics, intelligent
automation, and enhanced user experiences in various domains.
2.8 Biotechnology: Advances in biotechnology, including gene editing, syn-
thetic biology, and personalized medicine, hold promise for addressing
pressing global challenges in healthcare, agriculture, and environmental
sustainability. Ethical considerations and regulatory frameworks remain
important factors in biotech innovation.
2.9 Digital Twins: Digital twin technologies, which create virtual replicas of
physical assets, have seen adoption across industries for predictive main-
tenance, simulation, and optimization. Digital twins enhance operational
efficiency and decision-making by providing real-time insights into asset
performance.
2.10 Augmented Reality and Virtual Reality: AR and VR applications con-
tinue to expand beyond entertainment, with use cases in training, educa-
tion, healthcare, and remote collaboration. Immersive experiences offer
new opportunities for engagement and interaction.
Exploring Future Trends and Emerging Applications 201

The Gartner hype cycle for emerging technologies in 2023 reflects a dynamic land-
scape of innovation and disruption with technologies at various stages of maturity
and adoption. By understanding these trends, organizations can strategize effec-
tively and harness the potential of emerging technologies to drive growth and
innovation.

PROPOSED METHODS FOR VARIOUS FUTURE

TRENDS AND EMERGING APPLICATIONS
The evolving landscape of technology unveils future trends and emerging appli-
cations across various domains, reshaping how we live, work, and interact.
Artificial intelligence is a key driver, permeating sectors like healthcare, finance,
and education, promising automation, efficiency, and personalized experiences.
Healthcare sees advancements in telemedicine and personalized medicine, while
finance experiences disruption through blockchain and decentralized finance.
Transportation transitions to electric and autonomous vehicles, while education
adopts immersive technologies for enhanced learning. These innovations offer
both challenges and opportunities, promising a more connected, efficient, and
technologically enriched future. Adapting to these trends is crucial for individu-
als, businesses, and policymakers to harness their benefits effectively. Figure 11.2
demonstrates the overall emerging and future trends of recent technologies in
various domains.
Various types of future and emerging trends have been briefly discussed as
follows:

FIGURE 11.2 Schematic framework of emerging and future trends of recent technologies.
202 Artificial Intelligence Using Federated Learning

Edge Computing
Edge computing is a paradigm in which data processing and storage are con-
ducted closer to the data source, typically at the edge of the network, rather than
relying solely on centralized data centers. This proximity to the data source
reduces latency, bandwidth usage, and reliance on cloud services, making it par-
ticularly advantageous for applications that require real-time or near-real-time
processing. The emergence of the Internet of Things, coupled with advancements
in robust cloud services, has paved the way for a groundbreaking computing con-
cept termed edge computing [34]. Operating on both downstream data for cloud
services and upstream data for IoT services, edge computing introduces fresh
opportunities across diverse application domains [35]. It empowers the handling
of vast datasets generated at the IoT edge and optimizes the implementation of
ML, deep learning, and augmented discovery of data in harmony with the genera-
tion of data. Through a gradual reduction in cost and complexity within the cloud,
edge computing facilitates the seamless integration of deep learning functional-
ities into applications [36].

Key Characteristics and Components of Edge Computing

1. Proximity to Data Source: Edge computing infrastructure is deployed
closer to where data is generated or consumed. This could be in locations
such as factory floors, retail stores, vehicles, or IoT devices. By processing
data at the edge, latency is minimized, and bandwidth requirements are
reduced, as data doesn’t need to travel long distances to centralized data
centers.
2. Distributed Architecture: Edge computing relies on a distributed architec-
ture, where computing resources such as servers, storage, and networking
equipment are deployed at various edge locations. This allows for parallel
processing and efficient utilization of resources across the network.
3. Scalability: Edge computing architectures are designed to be scalable,
allowing organizations to easily deploy and manage edge nodes as needed.
This scalability is essential for accommodating growing data volumes and
processing requirements, especially in dynamic environments such as IoT
deployments.
4. Edge Devices: Edge devices, such as IoT sensors, actuators, and gateways,
play a crucial role in collecting and transmitting data to edge computing
nodes. These devices often have limited processing capabilities and rely
on edge computing infrastructure to perform more complex analysis and
decision-making.
5. Edge Computing Platforms: Edge computing platforms provide the nec-
essary software and tools to manage and orchestrate edge computing
resources effectively. These platforms may include edge analytics software,
security mechanisms, and management tools for monitoring and provision-
ing edge nodes.

Figure 11.3 depicts an architectural representation of edge computing techniques.

Exploring Future Trends and Emerging Applications 203

FIGURE 11.3 Architectural layer-wise representation of edge computing.

Internet of Things
The Internet of Things encompasses a web of interconnected devices, objects, and
systems that interact and exchange data through the internet. Equipped with sen-
sors, actuators, and connectivity capabilities, these devices gather and share data,
enabling automation, monitoring, and control of real-world environments. IoT
facilitates the networking of physical objects enabling sensors attached to these
objects to convey information regarding their status or surroundings. Authors in
a recent study [37] discussed key enabling technologies, indicating the driving
forces behind future IoT research endeavors. This connectivity opens a vast array
of applications across various domains including smart homes, smart cities, smart
healthcare, smart agriculture, and smart transportation [38]. The IoT holds vast
potential, with expected progress in resource control, energy management, qual-
ity of service, interoperability, and interface management as well as security and
privacy enhancements [39].

Key Components of the Internet of Things

1. Devices: IoT devices can range from simple sensors and actuators to com-
plex machines and appliances. These devices are embedded with technol-
ogy to collect data, perform actions, and communicate with other devices
or central systems.
2. Connectivity: Connectivity is essential for IoT devices to exchange data
with each other or with central systems such as cloud platforms or edge
servers. Common connectivity technologies used in the IoT include Zigbee;
Bluetooth; Wi-Fi; and cellular networks like 4G LTE, 4G, 5G, and satellite
communication.
3. Data Processing: The IoT generates vast amounts of data from connected
devices. Data processing techniques such as edge computing and cloud
204 Artificial Intelligence Using Federated Learning

computing are used to analyze and derive insights from this data. Edge
computing involves processing data locally on the device or at the edge of
the network, while cloud computing involves processing data in remote data
centers.
4. Applications and Services: IoT applications and services leverage data
collected from connected devices to offer various functionalities, includ-
ing smart home automation, industrial automation, healthcare monitoring,
environmental monitoring, smart city solutions, asset tracking, and predic-
tive maintenance.
5. Security and Privacy: Security and privacy are critical considerations in
IoT deployments to protect sensitive data and prevent unauthorized access
to devices and networks. Security measures such as encryption, authentica-
tion, access control, and device management are implemented to mitigate
cybersecurity risks in IoT systems.

Examples of IoT applications and use cases include smart homes, healthcare,
and smart cities. The Internet of Things has the potential to transform industries,
improve quality of life, and drive innovation by connecting the physical and digital
worlds in ways never before possible. As IoT adoption continues to grow, it will
play an increasingly significant role in shaping the future of technology and society.
Figure 11.4 showcases the process of an IoT application.

Virtual Reality
Virtual reality is an immersive technology that simulates a realistic, three-dimen-
sional environment using computer-generated imagery and interactive experiences.

FIGURE 11.4 IoT application processes.

Exploring Future Trends and Emerging Applications 205

Through VR, users can interact with and explore virtual environments as if they
were physically present within them. This technology typically involves the use of
a VR headset or goggles, along with motion-tracking sensors and handheld con-
trollers, to enhance the sense of presence and interaction. Virtual reality systems
facilitate situated learning by immersing users in interactive environments, objects,
and processes. These systems prioritize human interactions and engagement with
virtual experiences and environments, fostering learning through hands-on, interac-
tive experiences [40].

Key Components and Features of Virtual Reality

1. Headsets: VR headsets are worn over the eyes and display stereoscopic
images to create the illusion of depth and immersion. These headsets may
be tethered to a powerful computer or console, or they may be standalone
devices with built-in processing capabilities.
2. Motion Tracking: VR systems often utilize motion-tracking sensors to
monitor the user’s movements and adjust the virtual environment accord-
ingly. This allows users to interact with virtual objects and navigate through
virtual spaces naturally and intuitively.
3. Controllers: Handheld controllers are commonly used in VR to provide
users with a means of interacting with virtual objects and manipulating the
virtual environment. These controllers may feature buttons, triggers, and
joysticks for input, and they may also incorporate motion-sensing technol-
ogy for enhanced interaction.
4. Immersive Environments: Virtual reality environments can range from
realistic simulations of real-world locations to fantastical and surreal worlds
limited only by the imagination of the creators. These environments may be
static or dynamic, allowing users to explore, interact and even create within
them.
5. Applications: VR has applications across a wide range of industries and
fields including gaming, entertainment, education, training, healthcare,
architecture, and design. In gaming and entertainment, VR provides play-
ers with immersive experiences that transport them into the virtual worlds
of their favorite games and movies. In education and training, VR enables
realistic simulations and hands-on learning experiences in fields such as
medicine, engineering, and aviation. In healthcare, VR can be used for pain
management, rehabilitation, and therapy, while in architecture and design,
it allows architects and designers to visualize and explore virtual spaces
before they are built.

Virtual reality is a powerful and versatile technology that has the potential to revolu-
tionize how we experience and interact with digital content and the world around us.
VR can revolutionize enterprises and improve people’s lives, and this ability is only
likely to increase as virtual reality gear becomes more widely available and inex-
pensive and as innovators keep pushing the frontiers of what can be accomplished in
virtual worlds. Figure 11.5 depicts an architectural diagram of virtual reality.
206 Artificial Intelligence Using Federated Learning

FIGURE 11.5 Architectural block diagram of VR.

Commercial UAVs
Commercial unmanned aerial vehicles (UAVs), commonly referred to as drones,
have witnessed a rapid rise in adoption and application across various industries
in recent years. These versatile flying machines, equipped with sensors, cameras,
and other payloads, offer numerous advantages in terms of data collection, surveil-
lance, inspection, and delivery. They have become essential in many fields, including
science, technology, and society, offering a broad variety of uses, including mili-
tary tasks, recreational activities, and monitoring. They have effects on safety and
confidentiality in both private and public spheres [41]. However, their adaptability
is accompanied by constraints, including limited range and power restrictions. To
ensure continuous and uninterrupted operation, these devices necessitate periodic
recharging at designated intervals [42].

Key Features and Applications of Commercial UAVs

1. Data Collection and Mapping: UAVs are utilized for aerial surveys, map-
ping, and photogrammetry, providing high-resolution imagery and geo-
spatial data for urban planning, agriculture, forestry, and environmental
monitoring.
2. Surveillance and Security: UAVs are employed for surveillance and secu-
rity purposes in sectors such as law enforcement, border patrol, and infra-
structure protection, enabling real-time monitoring of critical areas and
rapid response to incidents.
3. Infrastructure Inspection: UAVs facilitate the inspection of infrastructure
such as power lines, pipelines, bridges, and buildings, allowing for effi-
cient and cost-effective assessment of structural integrity and maintenance
needs.
Exploring Future Trends and Emerging Applications 207

4. Agriculture and Crop Monitoring: In agriculture, UAVs are utilized for crop
monitoring, pest detection, and precision agriculture practices, enabling
farmers to optimize crop yields, reduce inputs, and enhance overall farm
management.
5. Search and Rescue: UAVs play a vital role in search and rescue operations,
providing aerial reconnaissance and thermal imaging capabilities to locate
missing persons or disaster survivors in remote or hazardous environments.
6. Delivery and Logistics: Companies are exploring the use of UAVs for the
delivery of goods and packages in the last mile, particularly in areas with
limited infrastructure or during emergency situations where traditional
delivery methods may be impractical.

Benefits of Commercial UAVs

1. Cost-Effectiveness: UAVs enable faster and more cost-effective data col-
lection and monitoring compared to traditional methods, such as manned
aircraft or ground-based surveys.
2. Safety: By removing the need for human operators to be physically pres-
ent in hazardous or inaccessible environments, UAVs improve safety and
reduce the risk of accidents and injuries.
3. Efficiency: UAVs can cover large areas quickly and efficiently, providing
timely and accurate data for decision-making and operational planning.
4. Flexibility: UAVs can be deployed in various industries and environments,
offering versatility and adaptability to diverse applications and tasks.

In conclusion, commercial UAVs represent a transformative technology with sig-

nificant potential to revolutionize various industries and sectors. As technology
advances and regulations evolve, the adoption and integration of UAVs into commer-
cial operations are expected to continue expanding unlocking new opportunities and
driving innovation across different domains. Figure 11.6 demonstrates a workflow
diagram of UAVs.

FIGURE 11.6 Workflow architecture of UAVs.

208 Artificial Intelligence Using Federated Learning

Software-Defined Networks
Software-defined networking (SDN) is an innovative approach to network manage-
ment that enhances the agility, flexibility, and programmability of traditional net-
work infrastructures. In SDN, the control plane, responsible for making decisions
about where to send traffic, is decoupled from the data plane, which is responsible for
forwarding the actual traffic. This decoupling is achieved through the implementa-
tion of a centralized controller that manages and directs network traffic. The adop-
tion of software-defined networking in the context of 5G has emerged as a solution
to address the limitations of hardware-based cellular architectures. Traditional hard-
ware-centric designs often lack flexibility and are constrained by closed systems.
SDN 5G adoption introduces a more efficient approach, enabling the creation of cen-
tralized network architectures with programmable capabilities spanning the entire
network [43]. In response to this need, the authors in [44] have proposed SoftAir, an
SDN architecture tailored specifically for 5G wireless systems. While promising,
this approach has raised security concerns, as discussed in [45]. However, there is an
anticipation that software-defined networks will play a significant role in the evolu-
tion of the mobile industry [46].

Key Components and Characteristics of Software-Defined Networks

1. Controller: In SDN, a central controller serves as the brain of the network
making global decisions about how traffic should be directed based on a
comprehensive view of the network. The controller communicates with
switches and routers in the network through southbound application pro-
gramming interfaces (APIs) to configure and manage their behavior.
2. Data Plane: The data plane, also known as the forwarding plane, remains
responsible for the actual transmission of network packets. Switches and
routers in the network forward traffic based on the instructions received
from the centralized controller.
3. OpenFlow Protocol: OpenFlow is a standardized communication protocol
between the SDN controller and the devices in the network’s data plane. It
enables the controller to communicate with and manage the flow control to
the networking devices.
4. Programmability: SDN allows network administrators and operators to
programmatically control and manage the network through software appli-
cations. This programmability facilitates automation, making it easier to
adapt the network to changing requirements.
5. Flexibility and Scalability: SDN enhances network flexibility, allowing
administrators to adjust network behavior via software without altering the
physical infrastructure. This architecture is highly scalable, making it suit-
able for dynamic and large-scale networks.

Applications and Benefits of Software-Defined Networks

1. Network Virtualization: SDN enables the creation of virtual network over-
lays, allowing multiple virtual networks to coexist on the same physical
network infrastructure.
Exploring Future Trends and Emerging Applications 209

2. Dynamic Traffic Management: With a centralized controller, SDN allows

for dynamic traffic management, optimizing routing decisions based on
real-time network conditions.
3. Automation and Orchestration: SDN facilitates automation of network pro-
visioning, configuration, and management tasks, improving operational
efficiency.
4. Improved Security: Centralized control provides enhanced visibility and
control over network traffic, enabling better security measures and threat
detection.
5. Efficient Resource Utilization: SDN enables more efficient use of network
resources by dynamically adapting to changing demands, reducing under-
utilization and congestion.

In summary, software-defined networking represents a paradigm shift in network

management, offering greater flexibility, control, and efficiency, particularly in the
face of evolving network requirements and technologies. Figure 11.7 demonstrates
the architecture of software-defined networks.

Artificial General Intelligence

AGI signifies the hypothetical capability of an artificial intelligence system to
perceive, learn, and implement knowledge like human intelligence throughout an
extensive array of tasks and domains. Unlike narrow AI, which is designed for spe-
cific tasks, AGI aims to possess human-like cognitive capabilities, including rea-
soning, problem-solving, perception, learning, and understanding natural language.
Artificial general intelligence finds extensive applications across various domains,

FIGURE 11.7 Block diagram of SDN architecture.

210 Artificial Intelligence Using Federated Learning

encompassing areas such as cognitive computing, commercial drones, deep learning,

deep reinforcement learning, smart robots, machine learning, autonomous vehicles,
user interfaces for conversation, smart dust, and smart workspaces. This pervasive
presence of AGI is often referred to as “artificial general intelligence everywhere”,
indicating its widespread integration into numerous high-tech services within cyber-
space [2].

Key Features and Characteristics of Artificial General Intelligence

1. Generalization: AGI systems can generalize knowledge and skills learned
in one domain to solve problems and tasks in different domains. This capa-
bility enables adaptability and versatility across a wide range of contexts.
2. Learning: AGI systems can learn from experience, acquiring new knowl-
edge and skills through observation, interaction, and feedback. This learn-
ing process may involve supervised learning, unsupervised learning,
reinforcement learning, or a combination of these approaches.
3. Reasoning and Problem-Solving: AGI systems can perform complex rea-
soning and problem-solving tasks, including logical deduction, inference,
planning, and decision-making. These abilities enable AGI to tackle novel
and challenging problems autonomously.
4. Natural Language Understanding: AGI systems can understand and gener-
ate human language in various forms, including speech, text, and dialogue.
This enables seamless communication and interaction with humans in nat-
ural language.
5. Perception: AGI systems can perceive and interpret sensory information
from the environment including visual, auditory, and tactile inputs. This
perception allows AGI to understand and interact with the physical world.
6. Autonomy: AGI systems possess a degree of autonomy, allowing them to
operate and make decisions independently without human intervention.
This autonomy enables AGI to adapt to changing circumstances and solve
problems in real time.

Challenges and Considerations in Artificial General Intelligence

1. Complexity: Achieving AGI entails addressing the complexity of human
intelligence which involves various cognitive processes, emotions, and
social interactions.
2. Ethical and Societal Implications: The development and deployment of AGI
raise ethical concerns related to privacy, autonomy, bias, fairness, account-
ability, and the potential impact on employment and society at large.
3. Safety and Control: Ensuring the safety and control of AGI systems is criti-
cal to prevent unintended consequences or harmful behaviors. Research
in AI safety focuses on developing safeguards and control mechanisms to
mitigate risks associated with AGI.
4. Interdisciplinary Nature: AGI research requires collaboration across mul-
tiple disciplines, including computer science, cognitive science, neurosci-
ence, philosophy, psychology, and ethics, to address the complexity of
human intelligence comprehensively.
Exploring Future Trends and Emerging Applications 211

FIGURE 11.8 Workflow diagram of AGI.

In the science of computational intelligence, general intelligence based on artifi-

cial intelligence is a revolutionary and inspirational aim that can completely change
many facets of business, society, and human existence. While achieving AGI
remains a long-term objective, ongoing research and advancements in AI continue
to push the boundaries of what is possible, bringing us closer to realizing the vision
of truly intelligent machines. Figure 11.8 shows a workflow diagram of artificial
general intelligence.

Deep Learning
Deep learning is a subfield of machine learning that involves using a data-driven
methods of learning to train neural networks made from data to do particular jobs.
Its notable rise stems from its capacity to autonomously discern and represent com-
plex patterns and features embedded within extensive datasets. Deep learning uti-
lizes back-propagation algorithms to uncover complex and intricate data structures
within big data. With the development of sophisticated convolutional neural networks
that have greatly improved the processing of voice, pictures, video and audio, this
strategy has been very successful. Furthermore, neural networks made up of recur-
rent neurons have proven to be highly effective in processing data that is sequential,
including voice and text [47]. Deep learning enables computers to learn from and
comprehend the world as a hierarchy of concepts. Unlike traditional methods, deep
learning minimizes the need for human intervention in knowledge acquisition, as
computers autonomously gather knowledge from their experiences. Through a hier-
archical structure of concepts, computers can effectively learn complex ideas from
simpler ones leveraging multiple deep layers to enhance their understanding [48, 49].
212 Artificial Intelligence Using Federated Learning

Key Aspects and Applications of Deep Learning

1. Neural Networks: Deep learning primarily relies on neural networks,
which are structured to mimic the human brain’s interconnected neurons. It
consists of multiple deep layers those enable the model to learn hierarchical
representations of data.
2. Image and Speech Recognition: Deep learning has revolutionized image
recognition, enabling accurate classification and detection such as recogni-
tion of object and facial recognition. In speech recognition, it has enhanced
the capabilities of virtual assistants and voice-controlled systems.
3. Natural Language Processing (NLP): Various tasks of NLP such as lan-
guage translation, sentiment analysis, and chatbots benefit from deep learn-
ing models that can understand and generate human-like language patterns.
Transformer models like BERT and GPT have significantly advanced NLP
applications.
4. Finance and Business: In the financial sector, deep learning is applied to
fraud detection, algorithmic trading, credit scoring, and customer service.
Businesses leverage deep learning for predictive analytics and recommen-
dation systems.
5. Generative Models: Generative models such as generative adversarial net-
works (GANs) can create new data samples that resemble existing datasets.
This is used in image generation, style transfer, and creating realistic syn-
thetic data.
6. Transfer Learning, Explainability and Interpretability: Transfer learning
enables models pre-trained on one task to be adapted for a different but
related task. This approach is efficient in scenarios with limited labeled
data. Addressing the “black-box” nature of deep neural networks, research
is ongoing to enhance model interpretability and explainability, making it
easier to understand the decisions made by these complex models.

Deep learning’s success can be attributed to the availability of large datasets, power-
ful computing resources, and advancements in model architectures. Deep Learning
is poised to drive further innovations and improvements in various applications, con-
tributing to the advancement of artificial intelligence.

Machine Learning
Machine learning is a dynamic field within the broader realm of artificial intel-
ligence that focuses on developing algorithms and models capable of enabling
computers to learn from data. Instead of relying on explicit programming, ML
systems utilize statistical techniques to improve their performance on a specific
task over time. The essence of machine learning lies in the ability of algorithms to
identify patterns, make predictions, or optimize outcomes based on the data they
are exposed to. There are various types of machine learning including supervised
learning, unsupervised learning, and reinforcement learning, each serving distinct
purposes.
Exploring Future Trends and Emerging Applications 213

Key Concepts and Components of Machine Learning

1. Data: Data is the foundation of machine learning. Algorithms learn from
patterns and information within datasets, which can include a wide range of
variables and features.
2. Algorithms: Machine learning algorithms are the mathematical models
that process data and learn patterns. Common types of algorithms include
supervised learning, unsupervised learning, and reinforcement learning.
3. Supervised Learning: In this method, the algorithm is trained using a
labelled dataset in which each input pair has an associated collection of out-
put labels. The examples given teach the algorithm how to translate inputs
data to the right output.
4. Unsupervised Learning: Using an unlabeled dataset, the algorithm is trained
by unsupervised learning. Among the data, the algorithm finds structures,
connections, or patterns without specific instructions on the output.
5. Reinforcement Learning: Reinforcement learning is a process where an
algorithm learns to make consecutive decisions within an environment to
attain a predetermined objective. Through this process, the algorithm is
provided with feedback either in the form of rewards or penalties based on
the actions it takes.
6. Feature Engineering: Feature engineering involves selecting and transform-
ing relevant features (variables) in the dataset to enhance the performance
of machine learning models.
7. Model Evaluation: Machine learning models need to be evaluated to ensure
their effectiveness. Common evaluation metrics include accuracy, preci-
sion, recall, and F1 score.
8. Training and Testing: Datasets are typically divided into training and test-
ing sets. The model is trained on the training set and then evaluated on the
testing set to assess its generalization performance.

Applications of Machine Learning

1. Image and Speech Recognition: Machine learning is widely used in image
and speech recognition systems, powering applications like facial recogni-
tion, virtual assistants, and language translation.
2. Finance: In the financial sector, machine learning is utilized for fraud
detection, credit scoring, algorithmic trading, and risk assessment.
3. Recommendation Systems: Online platforms use machine learning to build
recommendation systems that provide personalized suggestions for users
based on their preferences and behavior.
4. Autonomous Vehicles: Machine learning plays a crucial role in the develop-
ment of autonomous vehicles, enabling them to perceive and respond to the
surrounding environment.
5. Natural Language Processing: Machine learning powers natural language
processing applications, such as chatbots, sentiment analysis, and language
translation.
214 Artificial Intelligence Using Federated Learning

FIGURE 11.9 Representation of operational steps of machine learning.

As technology advances, machine learning continues to evolve, offering new pos-

sibilities for automation, prediction, and decision-making across various industries.
Figure 11.9 depicts the operational architecture of machine learning.

Digital Twins
A digital twin is a virtual representation of a physical object, system, or process. It
is created using real-time data and simulation models to mimic the behavior, char-
acteristics, and performance of its physical counterpart. Digital twins enable orga-
nizations to monitor, analyze, and optimize assets and operations throughout their
life cycle, providing valuable insights for decision-making and improving efficiency.

Key Components and Characteristics of Digital Twins

1. Virtual Representation: A digital twin replicates the physical attributes,
behavior, and interactions of its real-world counterpart in a digital environ-
ment. This virtual representation is created using sensor data, CAD models,
and other relevant information.
2. Real-Time Data Integration: Digital twins leverage real-time data from sen-
sors, IoT devices, and other sources to continuously update and synchronize
the virtual model with the physical object or system. This allows for accu-
rate monitoring and analysis of current conditions and performance.
3. Simulation and Analytics: Digital twins incorporate simulation and analyt-
ics capabilities to simulate scenarios, predict behavior, and analyze perfor-
mance. By running simulations and what-if scenarios, organizations can
assess the impact of changes and optimize operations.
4. Lifecycle Management: Digital twins support the entire lifecycle of assets
and systems from design and development to operation and maintenance.
Exploring Future Trends and Emerging Applications 215

They provide insights into asset performance, health, and maintenance

needs, enabling proactive decision-making and optimizing lifecycle costs.
5. Interconnectivity: Digital twins can be interconnected with other digital
twins, systems, and applications, forming a network of interconnected
assets and processes. This interconnectivity enables holistic analysis, col-
laboration, and optimization across multiple domains.
6. Applications Across Industries: Digital twins find applications across vari-
ous industries, including manufacturing, healthcare, transportation, energy,
and smart cities. They are used for predictive maintenance, product design
and optimization, supply chain management, healthcare simulation, and
urban planning, among others.
7. Predictive Maintenance: In manufacturing and asset-intensive industries,
digital twins enable predictive maintenance by monitoring equipment
health and performance in real time. By detecting anomalies and predict-
ing failures before they occur, organizations can minimize downtime and
maintenance costs.
8. Product Design and Optimization: In product development, digital twins
allow engineers to simulate and optimize designs, test performance under
different conditions, and iterate rapidly. This accelerates the product devel-
opment process and improves product quality.
9. Smart Cities and Infrastructure: In smart cities and infrastructure projects,
digital twins are used to model and simulate urban environments, infra-
structure systems, and transportation networks. This enables city planners
to optimize resources, improve resilience, and enhance citizen services.

Digital twins serve an essential function in digital transformation initiatives, enabling

organizations to gain deeper insights, improve decision-making, and drive innova-
tion across various domains, depicted in Figure 11.10. As technology advances and
data connectivity increases, the adoption of digital twins is expected to grow, leading
to further optimization and transformation of industries and processes.

FIGURE 11.10 Operational function of digital twins.

216 Artificial Intelligence Using Federated Learning

SERVERLESS PLATFORM-AS-A-SERVICE
Serverless Platform-as-a-Service represents a significant evolution in cloud com-
puting architecture, offering a hardware-free environment for orchestrating and
managing cloud-based processes. Unlike traditional approaches that rely on infra-
structure-based designs, serverless PaaS introduces a paradigm shift by abstracting
away the complexities of server management. At its core, serverless PaaS operates
on the concept of Function-as-a-Service (FaaS), where computing resources are pro-
vided on demand in response to specific events or triggers. This model eliminates the
need for developers to provision or manage servers, allowing them to focus solely on
writing and deploying code [2]. This approach enhances scalability, agility, and cost-
effectiveness, as resources are allocated dynamically based on workload demands.
One of the key advantages of serverless PaaS is its ability to streamline the develop-
ment and deployment of applications without the overhead of infrastructure manage-
ment. Developers can write code in the form of functions, which are executed in
response to events triggered by external sources such as HTTP requests, database
changes, or scheduled tasks. Serverless PaaS architectures offer inherent scalability,
as resources are provisioned automatically to accommodate fluctuating workloads.
This elasticity ensures optimal performance and cost efficiency, as organizations only
pay for the resources consumed during code execution. Security is another critical
aspect addressed by serverless PaaS. By abstracting away the underlying infrastruc-
ture, potential attack surfaces are minimized, reducing the risk of security breaches.
Moreover, serverless PaaS providers often offer built-in security features such as
encryption, access controls, and monitoring tools to further enhance the security
posture of applications. Serverless PaaS represents an approach to cloud computing
offering developers a hassle-free and cost-effective way to build and deploy applica-
tions at scale. As organizations continue to embrace digital transformation initia-
tives, serverless PaaS is poised to play a pivotal role in driving innovation and agility
in the ever-evolving landscape of cloud computing. Figure 11.11 showcases the PaaS
architecture in cloud computing.

FIGURE 11.11 Serverless PaaS architecture in cloud computing.

Exploring Future Trends and Emerging Applications 217

Cognitive Computing
At its core, cognitive computing is a multidisciplinary field dedicated to crafting
computational models and decision-making mechanisms, drawing inspiration from
the neurobiological processes of the human brain, cognitive sciences, and psychol-
ogy [50]. Unlike traditional computing systems, cognitive computing systems aim
to mimic human-like intelligence and learning capabilities enabling them to under-
stand, reason, and learn from data in a manner that resembles human cognition.
The key objective of cognitive computing is to create systems that can process vast
amounts of data, understand natural language, and make context-aware decisions.
This involves the utilization of techniques such as machine learning, pattern recogni-
tion, natural language processing, and computer vision. These systems continuously
learn and adapt, allowing them to improve their performance over time-based on
experience and new data. Cognitive computing systems excel in handling unstruc-
tured data, which is prevalent in the real world, such as text, images, and audio. By
leveraging advanced algorithms, these systems can extract meaningful insights, rec-
ognize patterns, and infer context from diverse data sources. One prominent applica-
tion of cognitive computing is in the development of virtual assistants, chatbots, and
conversational interfaces. These systems can understand and respond to user queries
in a natural language format, providing a more intuitive and user-friendly interac-
tion. In healthcare, cognitive computing contributes to diagnostic processes by ana-
lyzing vast datasets of medical records, images, and research papers. By identifying
patterns and correlations, these systems assist healthcare professionals in making
more accurate and timely decisions.
The technology also finds applications in finance, customer service, and vari-
ous industries where data-driven decision-making is crucial. Cognitive computing
systems are capable of sifting through enormous datasets, automating routine tasks,
and providing valuable insights for strategic planning. The development of cognitive
computing represents a significant leap forward in the evolution of artificial intel-
ligence. As technology continues to advance, the integration of cognitive computing
into various domains is poised to bring about progressive changes, enhancing effi-
ciency, decision-making, and user experiences across a spectrum of applications, as
shown in Figure 11.12.

Blockchain
Blockchain stands as a groundbreaking technology that functions as the fundamen-
tal mechanism behind cryptocurrencies like Bitcoin, but its applications extend far
beyond the realm of digital currencies. Blockchain is a decentralized and distributed
ledger that records transactions across a network of computers in a secure and trans-
parent manner. The primary feature of blockchain is its ability to provide a secure
and tamper-resistant way of recording transactions. In a blockchain, each transac-
tion is grouped into a block, and these blocks are linked together in a chronological
chain. Once a block is added to the chain, it becomes extremely difficult to alter
any information within it, ensuring the integrity of the entire transaction history.
Key characteristics of blockchain include decentralization, persistency, efficiency,
218 Artificial Intelligence Using Federated Learning

FIGURE 11.12 The application of cognitive computing in various domains.

secrecy, and auditability. Unlike traditional centralized databases, blockchain oper-

ates on a peer-to-peer network, where each participant (node) has a copy of the entire
ledger. This decentralized nature eliminates the need for a central authority and
enhances the security of the system [51].
The concept of decentralization in blockchain introduces trust and transparency.
Transactions are verified through a consensus mechanism, often referred to as min-
ing in the context of cryptocurrencies. This ensures that all participants in the net-
work agree on the validity of transactions, reducing the risk of fraud. Blockchain
technology finds applications beyond cryptocurrency. It is used for secure and trans-
parent transactions, reducing the reliance on traditional banking systems. Smart con-
tracts, self-executing contracts with the terms of the agreement directly written into
code, are another innovative use of blockchain technology to automate and enforce
contractual agreements. Industries such as supply chain management, healthcare,
and logistics leverage blockchain to enhance transparency, traceability, and security.
The immutability of the blockchain ensures that records are tamper-proof, providing
a reliable and transparent trail of transactions. Despite its innovative potential, block-
chain is not without challenges. Issues related to scalability, energy consumption in
certain consensus mechanisms, and regulatory concerns are areas that continue to
be explored and addressed. Nevertheless, the underlying principles of blockchain
have sparked a paradigm shift in how we envision trust, security, and decentralized
systems, making it one of the most groundbreaking technologies of the digital age.
Figure 11.13 shows different types of architecture of blockchain.

Human Augmentation
Human augmentation is a technological field focused on enhancing and extending
human capabilities, whether naturally or artificially. This emerging area explores
Exploring Future Trends and Emerging Applications 219

FIGURE 11.13 Various types of blockchain architecture.

various ways to overcome the limitations of the human body, pushing boundaries
and augmenting functionalities. The primary objective is to improve and expand
human performance, cognition, and physical attributes through the integration of
advanced technologies. Technological interventions in human augmentation include
the use of wearable devices, implantable technologies, neuroprosthetics, and genetic
enhancements. These advancements aim to not only restore impaired functions but
also to enhance abilities beyond typical human capacities. Human augmentation
technologies often find applications in healthcare, accessibility, sports, and general
well-being. From prosthetic limbs with advanced sensory feedback to brain-machine
interfaces that enable direct communication between the brain and external devices,
human augmentation is at the forefront of merging biology with technology. The
ethical considerations surrounding privacy, consent, and the potential for societal
disparities are critical aspects that accompany the rapid progress in human augmen-
tation. As the field continues to evolve, researchers and innovators strive to strike a
balance between the benefits of improved human capabilities and the ethical impli-
cations associated with modifying the essence of what it means to be human. The
pursuit of responsible and ethical practices in human augmentation remains piv-
otal as society navigates this disruptive intersection of biology and technology [51].
Figure 11.14 demonstrates various functions of human augmentation.

Augmented Reality
Augmented reality is a cutting-edge technology that overlays computer-generated
information onto the real-world environment, providing users with an enhanced
and interactive perception of their surroundings as depicted in Figure 11.15. Unlike
virtual reality, which immerses users in a completely simulated environment, aug-
mented reality supplements the physical world with digital elements in real time.
Key features of augmented reality include the integration of computer-generated
sensory input such as sound, video, graphics, or GPS data into the user’s real-time
environment. AR applications are diverse and span various sectors, including gam-
ing, education, healthcare, manufacturing, and navigation. AR enhances user expe-
riences by blending the digital and physical realms, offering a seamless interaction
220 Artificial Intelligence Using Federated Learning

FIGURE 11.14 Functions of human augmentation.

between the virtual and real worlds. This technology can be experienced through
specialized AR glasses, smartphones, tablets, or other devices equipped with cam-
eras and sensors [52].
In the educational sector, AR can bring learning materials to life, offering inter-
active and immersive experiences. In healthcare, it aids in medical training, surgery
planning, and patient education. The gaming industry utilizes AR to create engaging
and interactive gameplay experiences by integrating virtual elements into the real
world. The continuous advancement of AR technology is driven by improvements in
hardware, such as better sensors and optics, and sophisticated software development.
As AR becomes more widespread, it is expected to revolutionize industries, redefine
how users engage with information, and create new opportunities for innovation
and collaboration. Despite its evolutionary potential, challenges such as privacy con-
cerns, technological limitations, and the need for seamless integration into everyday
life must be addressed. As AR continues to evolve, it holds the promise of reshaping
how individuals perceive and interact with their environment, unlocking new pos-
sibilities for communication, entertainment, and problem-solving.

Quantum Computing
Quantum computing represents a revolutionary approach to computation that har-
nesses the principles of quantum mechanics to perform operations on data in fun-
damentally new ways. Unlike classical computers, which process data using bits
Exploring Future Trends and Emerging Applications 221

FIGURE 11.15 Functional blueprint for integrating digital content into the real-time
environment.

represented as either 0s or 1s, quantum computers use quantum bits or qubits [51],
which can exist in multiple states simultaneously due to the principle of superposition.
One of the most remarkable features of quantum computing is its ability to perform
parallel computations on a vast number of possibilities simultaneously. This parallel-
ism allows quantum computers to solve certain problems much more efficiently than
classical computers, especially those involving complex calculations or vast amounts
of data. Another key principle of quantum computing is entanglement, which enables
qubits to become interconnected in such a way that the state of one qubit instantly
influences the state of another, regardless of the distance between them. This phe-
nomenon allows quantum computers to achieve a higher level of computational power
and efficiency. With the capacity to reshape multiple sectors such as drug discovery,
cryptography, artificial intelligence, materials science, and optimization, quantum
computing poses a notable impact. One instance is its potential to dismantle con-
ventional cryptographic algorithms by quickly factoring large numbers, leading to
the development of more secure encryption methods. Despite its immense potential,
quantum computing is still in its infancy, and many technical challenges remain to be
addressed. These challenges include qubit coherence [51] and stability, error correc-
tion, scalability, and the development of practical quantum algorithms.
Several companies, research institutions, and governments around the world are
investing heavily in quantum computing research and development. Major players in
the field include IBM; Google; Microsoft; Intel; and academic institutions such as
MIT, Caltech, and the University of Waterloo. As the field of quantum computing
continues to advance, it holds the promise of unlocking new frontiers in computa-
tion, enabling breakthroughs that were previously thought impossible. While prac-
tical quantum computers capable of outperforming classical computers on a wide
range of tasks are still years away, the progress being made in this field is both
exciting and transformative paving the way for a new era of computing. Figure 11.16
demonstrates the detailed architecture of the quantum computing technique.

Software-Defined Security
Software-defined security (SDS) is a paradigm in cybersecurity that emphasizes the
use of software-based security mechanisms to protect computer systems, networks,
222 Artificial Intelligence Using Federated Learning

FIGURE 11.16 Schematic diagram of detailed quantum computing process.

and data. Unlike traditional security approaches that rely heavily on hardware appli-
ances and manual configurations, SDS leverages software-defined networking prin-
ciples to provide a more flexible, scalable, and centralized security solution [53].
At its core, SDS separates the control plane from the data plane, allowing security
policies and configurations to be managed centrally through software rather than
being tied to specific hardware devices. This decoupling of control enables organi-
zations to dynamically adapt their security posture in response to evolving threats
and business requirements. Software-defined security represents a paradigm shift
in cybersecurity, offering organizations greater agility, scalability, and resilience in
the face of evolving cyber threats. By embracing SDS principles, organizations can
enhance their security posture and better protect their digital assets in an increas-
ingly complex and dynamic threat landscape. The principles of SDS are demon-
strated in Figure 11.17.

Conclusion and Future Scope

The exploration of future trends and emerging applications offers an intriguing
glimpse into the landscape that awaits humanity tomorrow. As society navigates
the ever-evolving realm of technology, it stands on the brink of progressive inno-
vations poised to reshape lifestyles, work environments, and social interactions.
The multidisciplinary nature of cognitive computing, the decentralized and trans-
parent capabilities of blockchain, the potential for enhancing human capabilities
through augmentation, the immersive experiences offered by augmented reality,
and the revolutionary possibilities of quantum computing collectively underscore
the dynamic trajectory of the technological future. These advancements not only
Exploring Future Trends and Emerging Applications 223

FIGURE 11.17 Software-defined security paradigm.

push the boundaries of what was previously thought possible but also create new
opportunities for solving complex problems and addressing challenges across vari-
ous domains. Furthermore, the paradigm shift towards software-defined security
reflects the industry’s dedication to crafting flexible, centralized, and efficient solu-
tions to safeguard the increasingly interconnected digital world. The integration of
such technologies represents more than a mere compilation of isolated advance-
ments; it signifies the convergence of various disciplines, leading to a comprehensive
transformation that transcends individual sectors.
Looking ahead, the rise of autonomous vehicles, smart workspaces, efficient
hardware, and virtual assistants further cements the progression towards a more
intelligent and interconnected world. These trends hold the promise of enhancing
efficiency, convenience, and innovation across diverse industries, paving the way for
a future where technology seamlessly integrates into daily life. In this exploration
of the future, it becomes clear that staying attuned to emerging trends is not merely
an option but a necessity. Adapting to these changes demands active participation,
continuous learning, and a collective effort to responsibly harness the benefits they
offer. As society stands at the threshold of tomorrow’s landscape, the choices made
today will shape the trajectory of the digital future. Embracing these advancements
with a thoughtful and proactive approach will undoubtedly contribute to a more con-
nected, efficient, and technologically enriched world.
224 Artificial Intelligence Using Federated Learning

REFERENCES
[1] Lu, Y. (2017). Industry 4.0: A Survey on Technologies, Applications, and Open
Research Issues. Journal of Industrial Information Integration, 6, 1–10.
[2] Hahanov, V., Gharibi, W., Man, K.L., Iemelianov, I., Liubarskyi, M., Abdullayev, V., &
Chumachenko, S. (2018). Cyber-Physical Technologies: Hype Cycle 2017. In Cyber
Physical Computing for IoT-driven Services, 259–272. Springer, Cham. https://doi.
org/10.1007/978-3-319-54825-8_14
[3] Matheson, E., Minto, R., Zampieri, E.G.G., Faccio, M., & Rosati, G. (2019). Human–
Robot Collaboration in Manufacturing Applications: A Review. Robotics, 8, 100.
[4] Vijayakuymar, G., & Suresh, B. (2022). Significance and Application of Robotics in the
Healthcare and Medical Field. Transaction on Biomedical Engineering Applications
and Healthcare, 3, 13–18.
[5] Oliveira, L.F.P., Moreira, A.P., & Silva, M.F. (2021). Advances in Agriculture Robotics:
A State-of-the-Art Review and Challenges Ahead. Robotics, 10, 52.
[6] Lee, H.-Y., & Murray, C.C. (2019). Robotics in Order Picking: Evaluating Warehouse
Layouts for Pick, Place, and Transport Vehicle Routing Systems. International Journal
of Production Research, 57, 5821–5841.
[7] Alatise, M.B., & Hancke, G.P. (2020). A Review on Challenges of Autonomous Mobile
Robot and Sensor Fusion Methods. IEEE Access, 8, 39830–39846.
[8] Soori, M., & Arezoo, B. (2023). Dastres, R. Artificial Intelligence, Machine Learning
and Deep Learning in Advanced Robotics, a Review. Cognitive Robotics, 3, 54–70.
[9] Raj, M., & Seamans, R. (2019). Primer on Artificial Intelligence and Robotics. Journal
of Organization Design, 8, 11.
[10] Mosavi, A., & Varkonyi-Koczy, A. (2017). Learning in Robotics. International Journal
of Computer Applications, 157, 975–8887.
[11] Van Den Hoven Van Genderen, R. (2017). Privacy and Data Protection in the Age of
Pervasive Technologies in AI and Robotics. European Data Protection Law Review, 3,
338–352.
[12] Elliott, D., & Soifer, E. (2022). AI Technologies, Privacy, and Security. Frontiers in
Artificial Intelligence, 5, 826737.
[13] Chiacchio, F., & Petropoulos, G. (2018). Pichler, D. The Impact of Industrial Robots
on EU Employment and Wages: A Local Labour Market Approach. Bruegel Working
Paper. Brussels, Belgium: Bruegel.
[14] Wu, C.-J., Raghavendra, R., Gupta, U., Acun, B., Ardalani, N., Maeng, K., Chang,
G., Behram, F.A., Huang, J., Bai, C. et al. (2021). Sustainable AI: Environmental
Implications, Challenges and Opportunities. arXiv, arXiv:2111.00364.
[15] Kamran, S., Farhat, I., Talha Mahboob, A., Gagandeep Kaur, A., Liton, D., Abdul
Ghaffar, K., Rimsha, I., Irum, S., & Afifah, R. (2020). The Impact of Artificial
Intelligence and Robotics on the Future Employment Opportunities. Trends in
Computer Science and Information Technology, 5, 50–54.
[16] Salau, A.O., Demilie, W.B., Akindadelo, A., & Eneh, J. (2022). Artificial Intelligence
Technologies: Applications, Threats, and Future Opportunities. In Proceedings of
the ACI’22: Workshop on Advances in Computation Intelligence, Its Concepts &
Applications at ISIC 2022, Savannah, GA, 17–19 May, pp. 265–273.
[17] Barfield, W. (2018). Liability for Autonomous and Artificially Intelligent Robots.
Paladyn, Journal of Behavioral Robotics, 9, 193–203.
[18] Zhu, Q., Rass, S., Dieber, B., & Vilches, V.M. (2021). Cybersecurity in Robotics:
Challenges, Quantitative Modeling, and Practice. Foundations and Trends® in Robotics,
9, 1–129.
Exploring Future Trends and Emerging Applications 225

[19] Desai, A., Qadeer, S., & Seshia, S.A. (2018). Programming Safe Robotics Systems:
Challenges and Advances. In Leveraging Applications of Formal Methods, Verification
and Validation. Verification, Proceedings of the 8th International Symposium, ISoLA
2018, Limassol, Cyprus, 5–9 November 2018, Margaria, T., Steffen, B., Eds. Cham,
Switzerland: Springer International Publishing, 2018, pp. 103–119.
[20] Robla-Gomez, S., Becerra, V.M., Llata, J.R., Gonzalez-Sarabia, E., Torre-Ferrero,
C., & Perez-Oria, J. (2017). Working Together: A Review on Safe Human-Robot
Collaboration in Industrial Environments. IEEE Access, 5, 26754–26773.
[21] Alpaydin, E. (2020). Introduction to Machine Learning. Cambridge, MA: MIT
Press.
[22] Eggers, J.P., & Kaul, A. (2018). Motivation and Ability? A Behavioral Perspective
on the Pursuit of Radical Invention in Multi-Technology Incumbents. Academy of
Management Journal, 61(1), 67–93. http://doi.org/10.5465/amj.2015.1123.
[23] George, G., Osinga, E.C., Lavie, D., & Scott, B.A. (2016). Big Data and Data Science
Methods for Management Research. Academy of Management Journal, 59(5), 1493–
1507. http://doi.org/10.5465/amj.2016.4005.
[24] Tonidandel, S., King, E.B., & Cortina, J.M. (Eds.). (2015). Big Data at Work: The Data
Science Revolution and Organizational Psychology (1st ed., 382). New York: Taylor &
Francis. https://doi.org/10.4324/9781315780504
[25] Barnes, C.M., Dang, C.T., Leavitt, K., Guarana, C.L., & Uhlmann, E.L. (2018). Archival
Data in Microorganizational Research: A Toolkit for Moving to a Broader Set of Topics.
Journal of Management, 44(4), 1453–1478. http://doi.org/10.1177/0149206315604188.
[26] Weick, K.E. (1992). Agenda Setting in Organizational Behavior: A Theory-
Focused Approach. Journal of Management Inquiry, 1(3), 171–182. http://doi.
org/10.1177/105649269213001.
[27] Carter, W. (2020). Putting Choice in the Spotlight to Advance Theory on Organizational
Adaptation to Technological Discontinuities. Organization Management Journal,
17(2). http://doi.org/10.1108/OMJ-04-2019-0720.
[28] Von Krogh, G. (2018). Artificial Intelligence in Organizations: New Opportunities for
Phenomenon Based Theorizing. Academy of Management Discoveries, 4(4). http://doi.
org/10.5465/amd.2018.0084.
[29] Makadok, R., Burton, R., & Barney, J. (2018). A Practical Guide for Making Theory
Contributions in Strategic Management. Strategic Management Journal, 39(6), 1530–
1545. http://doi.org/10.1002/smj.2789.
[30] Bamberger, P.A., & Pratt, M.G. (2010). Moving Forward by Looking Back:
Reclaiming Unconventional Research Contexts and Samples in Organizational
Scholarship. Academy of Management Journal, 53(4), 665–671. http://doi.org/10.5465/
amj.2010.52814357.
[31] Becker, W.J., Cropanzano, R., & Sanfey, A.G. (2011). Organizational Neuroscience:
Taking Organizational Theory Inside the Neural Black Box. Journal of Management,
37(4), 933–961. http://doi.org/10.1177/0149206311398955.
[32] Tenhiälä, A., Giluk, T.L., Kepes, S., Simon, C., Oh, I.S., & Kim, S. (2016). The
Research-Practice Gap in Human Resource Management: A Cross-Cultural Study.
Human Resource Management, 55(2), 179–200. http://doi.org/10.1002/hrm.21656.
[33] Aguinis, H., Boyd, B.K., Pierce, C.A., & Short, J.C. (2011). Walking New Avenues in
Management Research Methods and Theories: Bridging Micro and Macro Domains.
Journal of Management, 37(2), 395–403. http://doi.org/10.1177/0149206310382456.
[34] Shi, W., Cao, J., Zhang, Q., Li, Y., & Xu, L. (2016). Edge Computing: Vision and
Challenges. IEEE Internet of Things Journal, 3(5), 637–646. http://doi.org/10.1109/
JIOT.2016.2579198
226 Artificial Intelligence Using Federated Learning

[35] Shi, W., & Dustdar, S. (2016). The Promise of Edge Computing. Computer, 49(5), 78–
81. http://doi.org/10.1109/MC.2016.145
[36] Satyanarayanan, M. (2017). The Emergence of Edge Computing. Computer, 50(1), 30–
39. http://doi.org/10.1109/MC.2017.9
[37] Gubbi, J., Buyya, R., Marusic, S., & Palaniswami, M. (2013). Internet of Things (IoT):
A Vision, Architectural Elements, and Future Directions. Future Generation Computer
Systems, 29(7), 1645–1660.
[38] Singh, K.J., & Kapoor, D.S. (2017). Create Your Own Internet of Things: A Survey
of IoT Platforms. IEEE Consumer Electronics Magazine, 6(2), 57–68. http://doi.
org/10.1109/MCE.2016.2640718.
[39] Yaqoob, I., Ahmed, E., Hashem, I.A.T., Ahmed, A.I.A., Gani, A., Imran, M., &
Guizani, M. (2017). Internet of Things Architecture: Recent Advances, Taxonomy,
Requirements, and Open Challenges. IEEE Wireless Communications, 24(3), 10–16.
http://doi.org/10.1109/MWC.2017.1600421.
[40] Greenwald, S., Kulik, A., Kunert, A., Beck, S., Frohlich, B., Cobb, S.,. . .& Snyder, A.
(2017). Technology and Applications for Collaborative Learning in Virtual Reality.
https://repository.isls.org/handle/1/210
[41] Rao, B., Gopi, A.G., & Maione, R. (2016). The Societal Impact of Commercial Drones.
Technology in Society, 45, 83–90. https://doi.org/10.1016/j.techsoc.2016.02.009
[42] Hong, I., Kuby, M., & Murray, A. (2017). A Deviation Flow Refueling Location Model
for Continuous Space: A Commercial Drone Delivery System for Urban Areas. In
Advances in Geocomputation, 125–132. Cham: Springer.
[43] Akyildiz, I.F., Lin, S.C., & Wang, P. (2015). Wireless Software-Defined Networks
(W-SDNs) and Network Function Virtualization (NFV) for 5G Cellular Systems: An
Overview and Qualitative Evaluation. Computer Networks, 93, 66–79. https://doi.
org/10.1016/j.comnet.2015.10.013.
[44] Akyildiz, I.F., Wang, P., & Lin, S.C. (2015). SoftAir: A Software Defined Networking
Architecture for 5G Wireless Systems. Computer Networks, 85, 1–18. https://doi.
org/10.1016/j.comnet.2015.05.007.
[45] Chen, M., Qian, Y., Mao, S., Tang, W., & Yang, X. (2016). Software-Defined Mobile
Networks Security. Mobile Networks and Applications, 21(5), 729–743.
[46] Chen, T., Matinmikko, M., Chen, X., Zhou, X., & Ahokangas, P. (2015). Software Defined
Mobile Networks: Concept, Survey, and Research Directions. IEEE Communications
Magazine, 53(11), 126–133. http://doi.org/10.1109/MCOM.2015.7321981.
[47] LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep Learning. Nature, 521(7553), 436.
http://doi.org/10.1038/nature14539
[48] Kim, K.G. (2016). Book Review: Deep Learning. Healthcare Informatics Research,
22(4), 351–354.
[49] Schmidhuber, J. (2015). Deep Learning in Neural Networks: An Overview. Neural
Networks, 61, 85–117. https://doi.org/10.1016/j.neunet.2014.09.003.
[50] Gutierrez-Garcia, J.O., & López-Neri, E. (2015). Cognitive Computing: A Brief
Survey and Open Research Challenges. In Applied Computing and Information
Technology/2nd International Conference on Computational Science and Intelligence
(ACIT-CSI), 2015 3rd International Conference on, 328–333. IEEE. http://doi.
org/10.1109/ACIT-CSI.2015.64.
[51] Zheng, Z., Xie, S., Dai, H.-N., Chen, X., & Wang, H. (2018). Blockchain Challenges
and Opportunities: A Survey. International Journal of Web and Grid Services, 8
Inderscience Enterprises Ltd., 14(4), 352–375.
Exploring Future Trends and Emerging Applications 227

[52] Dunleavy, M., & Dede, C. (2014). Augmented Reality Teaching and Learning. In
Handbook of Research on Educational Communications and Technology (pp. 735–
745). Springer. https://doi.org/10.1007/978-1-4614-3185-5_59
[53] Al-Ayyoub, M., Jararweh, Y., Benkhelifa, E., Vouk, M., & Rindos, A. (2015).
Sdsecurity: A software-defined security experimental framework. In Communication
Workshop (ICCW), 2015 IEEE International Conference on, 1871–1876. IEEE. http://
doi.org/10.1109/ICCW.2015.7247453
 12 Securing Federated
Deep Learning
Privacy Risks and
Countermeasures
Atharva Saraf, Shaurya Sameer Talewar,
Susanta Das, Khushbu Trivedi, and
Ahmed A. Elngar

INTRODUCTION
Artificial intelligence has rapidly gained global acceptance, driving a surge in
demand for powerful ML models. ML’s ability to uncover hidden insights and detect
complex patterns has revolutionized scientific research and real-world applications
[1]. On the other hand, this pursuit frequently comes into conflict with growing
worries over the privacy and security of data. The act of centralizing large volumes
of sensitive data, such as medical records or financial transactions, leaves data vul-
nerable to the possibility of breaches and access violations by unauthorized parties.
Furthermore, tough data privacy legislation such as the General Data Protection
Regulation (GDPR), which is the EU’s data protection regulation and went into
effect on May 25, 2018, and is aimed at safeguarding the privacy and security of
EU citizens’ data [2], and the California Consumer Privacy Act (CCPA) further
restrict data collecting and sharing, which hinders the creation of novel artificial
intelligence solutions [2].
Under these laws, operators are strictly required to clearly explain user agree-
ments and are not allowed to trick or coerce consumers into giving up their privacy
rights. Users must give consent before the operator can train models, and they have
the right to delete their personal information. Additionally, network operators are
prohibited from releasing, modifying, or deleting the personal information of the
consumers they collect [3].
Federated learning comes as a groundbreaking approach that reshapes the land-
scape of collaborative learning. McMahan et al. define federated learning as a
“method facilitating collaborative model training between devices and servers with-
out the need to exchange raw data” [4, 5]. This is different from traditional methods,
which are dependent on centralized data storage. By using a decentralized learning
framework, we get remarkable benefits while ensuring the protection of individual

228 DOI: 10.1201/9781003482000-12

Securing Federated Deep Learning 229

TABLE 12.1
Notation
Symbol Description
K The total number of client participants in the federated learning process.
Ck The Kth client in the federated learning system, where k = 1, 2,. . . ., K.
Xk The dataset of features held by the client Ck.
Yk The dataset of labels corresponding to Xk for client Ck.
0k The global model parameters at iteration t.
E The number of local epochs each client Ck runs before updating the global
model.
The local version of the global model parameters at client Ck after local training at
0 kt
iteration t.
l The cross-entropy loss function.

F0 t ( xk ) The prediction made by the local model with parameters 0kt on the mini-batch xk
k

xk , yk A mini-batch of data (feature and labels) sampled from ( X k , Yk ) during local

training.
( t+1) The updated global model parameters after aggregating updates from all clients at
0 iteration t +1.
D The dataset consisting of N samples, where each sample is represented as (Xi, Yi).
N The total number of samples in the dataset D.
Xi The feature of the ith sample in the dataset D.
Yi The label corresponding to the ith sample in the dataset D.
Q The joint machine learning (ML) model parameters.
The loss function for vertical federated learning.
l (Q; D)

f (Q; Xi , Yi ) The loss function applied to the joint ML model Q using sample Xi and
label Yi .
Y (Q ) The regularization term applied to the joint ML model Q.
V The hyperparameter controlling the strength of the regularization term Y (Q) in the
loss function.

privacy. Unlike the traditional method of machine learning, which requires gather-
ing and storing the data in a central location/server for model training, the decen-
tralized method of federated learning does not need storage in a central server or
location, as shown in Figure 12.1 [3].
Centralized data storage has many disadvantages, including concerns regard-
ing data breaches and unauthorized access, which could lead to the compromise
of sensitive information and privacy. Scalability is also an issue: as the volume of
data grows, centralized infrastructure struggles to keep the high demand of training
complex models requesting vast datasets [6]. The advantages of federated learning
are schematically shown in Figure 12.2.
230 Artificial Intelligence Using Federated Learning

FIGURE 12.1 General federated learning framework.

FIGURE 12.2 Advantages of federated learning.

Benefits of Decentralized Method

• Enhanced Privacy: Individuals get control over their data, which reduces
threats to their privacy that relate to traditional data centralization. It is
possible to maintain the confidentiality of sensitive information on user
devices, which helps in building trust and transparency in the machine
learning process.
Securing Federated Deep Learning 231

• Unleashed Scalability: Federated learning solves the challenge of training

models in vast and distributed datasets that are located in different parts
of the world. This is accomplished by combining the collective power of a
large number of devices [7].
• Enhanced Security: With no centralized data storage, the attack surface
is reduced, which makes it more difficult for breaches to penetrate huge
information sets. Even if a single device is breached, only a small portion
of the data is exposed, which improves the overall security and resilience
of the system [7].

Organizations can participate in collaborative learning while following the legal

requirements because of the decentralized nature of federated learning, which is in
line with data privacy standards such as the General Data Protection Regulation and
the California Consumer Privacy Act [2].
Federated learning not only leverages the advantages of nature but also opens up
new avenues such as on-device and edge learning, collaborative social research, and
broader applications. [7].
As artificial intelligence increasingly interferes with our lives, data privacy and
security concerns are increasing. Traditional machine learning methods involve
centralizing data, which increases privacy risks. Federated learning offers a new
approach by decentralizing model training and allowing collaborative work by shar-
ing raw data. In this chapter, we dive into the fundamental workings of federated
learning and its potential to protect data while improving AI.

BASIC CONCEPTS OF FEDERATED LEARNING

Definition
Federated learning (FL) is a method of collaborative model training across various
decentralized data sources or devices, ensuring data privacy and security. In this
method, the raw/original data stays restricted on individual devices or servers, and
the computation for model training occurs locally on these data sources [8].
The key components and processes that constitute FL are as follows [8]:

3.1.1. Decentralized Data Sources: Different data sources like smartphones,

Internet of Things (IoT) devices, or remote servers have their own pri-
vate datasets, often containing sensitive information. This prevents
direct data sharing.
3.1.2. Global Model Initialization: The process begins with setting up a global
machine learning model, typically a pre-trained neural network, as the
starting point for collaboration.
3.1.3. Local Model Training: Each data source trains its model locally using
its data and the global model as a reference.
3.1.4. Local Model Updates: After training, each data source calculates
updates based on the difference between its local model and the global
model. These updates reflect the knowledge gained from the local data.
232 Artificial Intelligence Using Federated Learning

3.1.5. Model Aggregation: Updates from all the data sources are securely
combined to create an updated global model. Techniques like advanced
cryptographic methods protect individual update privacy.
3.1.6. Model Deployment: The updated global model is deployed back to all
participating data sources, refining their models without revealing local
data.
3.1.7. Iterative Collaboration: This iterative process involves multiple rounds
of training, updates, and aggregation until a satisfactory model is
achieved.

classification of federated learning

Federated learning methods are usually subdivided into horizontal federated learning
(HFL), vertical federated learning (VFL), and federated transfer learning (FTL) [9].

Horizontal Federated Learning

Horizontal federated learning involves combining samples. It’s applicable when
there’s a significant overlap of participant data features but a small overlap of user
data. The data used for joint model training is where both parties share the same
data characteristics but have different users. Horizontal federated learning has broad
application scenarios, as shown in Figure 12.3 [9].

HFL Problem
The most popular horizontal federated learning method is FedAvg. Consider K cli-
K
ents {Ck }k=1, each with dataset ( X k , Yk ), where each X k has the same feature space.
At iteration t in FedAvg, the server sends a global model with parameters 0 t to all
clients. Each client Ck updates 0 t for given local epochs E to obtain the local ver-
(
sion of the global model 0kt by minimizing the cross-entropy loss L F0 t ( xk ), yk
k
)

FIGURE 12.3 Horizontal federated learning.

Securing Federated Deep Learning 233

evaluated at mini-batch ( xk , yk ) of ( X k , Yk ). Then it returns 0kt to the server that will

aggregate each 0kt to obtain 0 t+1 as [10]:

E0 .
1
0 t +1 = t
k (1)
K k =1

Vertical Federated Learning

Vertical federated learning involves combining features across users in different for-
mats. It’s applicable when there’s more overlap of users but fewer overlaps of data
features. The data used for joint model training is the part where data characteristics
for the same users on both sides are not identical, as shown in Figure 12.4 [9].

3.2.2.1. VFL Problem

A vertical federated learning system aims to collaboratively train a joint machine
N
learning model using a dataset D = {( Xi , Yi )} with N samples while preserving
i=1
the privacy and safety of local data and models. We formulate the loss of vertical
federated learning as follows.

1 n
E
K
min l (Θ; D ) V
Θ N E i =1
f (Θ; Xi , Yi ) + T
k =1
Y (Θ) (2)

where 0 denotes the joint ML model, f (.) and y (.) denote the loss and regularize
functions, and T is the hyperparameter that controls the strength of Y [11, 12].

Federated Transfer Learning

Federated transfer learning becomes relevant when there’s limited overlap in features
and samples among participants, like collaboration between banks and supermarkets

FIGURE 12.4 Vertical federated learning.

234 Artificial Intelligence Using Federated Learning

FIGURE 12.5 Federated transfer learning.

TABLE 12.2
Comparison of Three Federated Learning Frameworks
Federated Learning User Overlap Data Overlap Application Scenarios (Taking
Types Banks and Supermarkets as
Examples)
Horizontal federated Less Less Between banks in different regions
learning
Vertical federated Many Many Between banks and supermarkets
learning in same area
Federated transfer Less Less Between banks and supermarkets
learning in different areas

across different regions. This approach is primarily used in deep neural network sce-
narios to address challenges related to insufficient data or labels during joint model-
ing, as shown in Figure 12.5 [9].
Table 12.2 shows a comparison of three federated learning frameworks.

Federated vs Traditional Learning

Federated learning offers several advantages over traditional machine learning; see
Table 12.3.

3.3.1. Enhanced User Privacy: Federated learning ensures user data privacy
by storing data locally, without sharing it among participants [13]. This
approach, in conjunction with the data privacy, mandates outlined in the
General Data Protection Regulation [2].
3.3.2. Adaptability to Large-Scale Data: Leveraging large-scale training
data enhances model quality. Federated learning achieves better accu-
racy compared to traditional methods while also reducing the need for
Securing Federated Deep Learning 235

TABLE 12.3
Comparison Between Federated Learning, Traditional Machine Learning and
Deep Learning
Federated Traditional Machine Traditional Deep
Learning Learning Learning
Safety High Low Low
Amount of data Large Small or medium Small or medium
Sharing High None None
Model building process Multi-party Single-party Single-party

extensive equipment during training and speeding up model training

with large datasets [13].
3.3.3. Increased Flexibility of Data Sources: Federated learning enables
the inclusion of data sources that were previously unable to participate
due to specific constraints. These sources can now store data locally
while contributing to overall model training, thereby improving model
generalization.

LITERATURE REVIEW
In this literature review, we dived into the advantages, disadvantages, and practi-
cal implementations of federated learning in various domains, with a particular
focus on its applications in medical image analysis, personalized solutions, and the
integration of blockchain technology to enhance privacy and security, as shown in
Table 12.4.

DECENTRALIZED LEARNING
Decentralized training is the concept that lies at the core of the new approach that
federated learning takes. The usual paradigm of centralizing data for training is
overturned by this innovation, which instead gives devices and servers the ability
to learn together without revealing any of their sensitive information between them-
selves. According to Yang et al. [7], this change introduces several distinct issues that
are worth investigating, in addition to offering compelling advantages.

Protection of Personal Information

Privacy concerns are raised when data is centralized because it provides a honeypot
for potential attackers. Federated learning reduces the likelihood of these dangers
occurring by storing data either on the device itself or on local servers, ensuring that
information is never exposed to the outside world [1, 5, 15–17]. The only updates
to the model that are shared for aggregation are those that are produced from local
training on the data. Imagine that a language model is being trained on millions of
236 Artificial Intelligence Using Federated Learning
TABLE 12.4 (Continued)

Securing Federated Deep Learning

Advantages, Disadvantages, and Practical Implementations of Federated Learning in Various Domains
S.no References Advantages Disadvantages Practical Implementations
4 [17] • FL generates robust models without sharing • Imperfect training data leads to • Provides insights on federated learning–
data, ensuring privacy and security. highly inaccurate models. enabling technologies, protocols, and
• FL enables training across multiple devices • Data skewing issues due to applications.
without exchanging actual data. factors like imbalance and • Highlights challenges and advantages of
• FL provides higher security and access missing classes. federated learning systems.
privileges to data. • Communication challenges, • Offers a framework for personalized models
slower due to vast device in healthcare using federated learning.
quantity. • Explores the potential and applicability of
• Privacy concerns over raw data federated learning in various industries.
transmission in federated
learning.
5 [5] • Scalable production system for federated • Potential bias due to device • Addresses device availability, connectivity,
learning on mobile devices. training restrictions and storage, and computer resource challenges.
• Enables model training on a large corpus of deployment limitations. • Focuses on reducing bandwidth through
decentralized data. compression techniques and quantized
representation.
• Evaluates models in live AB experiments to
detect bias issues.
6 [11] • VF2Boost is 12.8–18.9× faster than existing • Sequential key procedures lead • Addresses device availability, connectivity,
implementations. to frequent idle waiting periods. storage, and computer resource challenges.
• Supports much larger datasets with tens of • Cryptography operations are • Focuses on reducing bandwidth through
millions of instances. time consuming and dominate compression techniques and quantized
• Customized operations speed up cryptography overall training time. representation.
operations in the system. • Evaluate models in live AB experiments to
• Achieves the same model accuracy as detect bias issues.
non-federated training.

237
• Employs a concurrent training protocol to
reduce idle periods.
(Continued )
TABLE 12.4 (Continued)
Advantages, Disadvantages, and Practical Implementations of Federated Learning in Various Domains

238
S.no References Advantages Disadvantages Practical Implementations
7 [18] • Addressing communication overhead in • Lack of fully secure • VF 2 Boost enhances vertical federated
federal training for high-efficiency demand. transmission may lead to learning speed significantly.
privacy concerns. • Enables cross-enterprise collaboration with
• Insider and outsider adversaries privacy guarantees.
pose threads to privacy during • Optimizes training workflow, supports larger
training. datasets, and speeds up operations.
• Improves CPU utilization and reduces public
network transmission.
8 [6] • FATE is an industrial-grade platform for • Limited scalability due to • Addresses data privacy and sharing concerns
collaborative learning with data protection. computational heaviness and in federated learning.
• Supports secure computation protocols, machine communication requirements. • Highlights the need for communication-

Artificial Intelligence Using Federated Learning

learning algorithms, and visualization tools. • May require significant efficient methods in the medical industry.
• Developed to bridge data silos, build models, coordination costs for • Discusses privacy risks related to indirect
and protect user privacy. large-scale training. privacy leakage during training.
9 [19] • Improved privacy, localized data processing, • Tensor Flow Lite imposes • FATE enables secure collaborative machine
and real-time decision-making capabilities. model size limitations for learning in industrial applications.
• Tenser Flow Lite benefits: improved mobile device deployment. • Supports secure computation protocols and
performance, reduced model size, offline • Limited support for custom machine learning algorithms for enterprises.
inference. operations in Tensor Flow Lite
can be challenging.
10 [3] • User privacy protection is ensured by locally • Data poisoning and model • Enhances emergency management apps on
stored federated learning data. poisoning are significant Android for security and efficiency.
• Model training adapted to large-scale data for disadvantages of federated • Facilities commercialization in data
improved training quality. learning. analytics, machine learning, and emergency
• Federated learning faces management.
security hazards from malicious • Offers potential for socialized data security
users and criminals. solutions and consultation services.
• Contributes to federated learning theory
with broad commercialization opportunities.

(Continued )
TABLE 12.4 (Continued)

Securing Federated Deep Learning

Advantages, Disadvantages, and Practical Implementations of Federated Learning in Various Domains
S.no References Advantages Disadvantages Practical Implementations
11 [1] • Federated learning offers privacy protection • The simple averaging method • Highlights advantages and pitfalls of
and handles data islands effectively. has limitations with low-quality federated learning in various industries.
• Blockchain integration enhances security, or malicious models. • Discusses security threads and protection
decentralization, and traceability in FL. • Communication bottlenecks measures in federated learning applications.
slow down training progress due • Addresses poisoning attacks in federated
to limited network bandwidth. learning models.
12 [7] • Enables privacy-preserving auctions, private • Enhances understanding of model
DNA comparisons, and threshold aggregation techniques in federated learning.
cryptography. • Provides insights for researchers to develop
• Can run machine learning models on data new aggregation techniques.
without revealing the model. • Assists practitioners in selecting appropriate
• Provides clean abstraction for system security aggregation methods for FL applications.
without understanding MPC protocols. • Categorizes aggregation techniques into
synchronous, hierarchical, and robust forms.
• Explores benefits of model aggregation in
smart healthcare and transportation.
13 [20] • FL offers a privacy guarantee compared to • GDPR requirements pose • MPC protocols are used in real-world
traditional ML approaches. challenges for conventional applications.
• FL systems mitigate the risk of data breaches cloud-centric ML approaches. • Ideal model simplifies MPC usage without
on centralized systems. • ML algorithms face constraints deep understanding.
• Privacy-preserving techniques in FL include in complying with GDPR’s
SMC and differential privacy. purpose limitation.
14 [2] • Federated learning protects privacy by • An unstable network environment • Addresses privacy in healthcare, wireless
dispersing training data. may lead to uploading communication, and service recommendations.
• FL allows for customized model training low-quality models in FL. • Discuss challenges like privacy protection,
without compromising user privacy. • Statistical heterogeneity can communication cost, and system heterogeneity.

239
impact FL due to varied data • Provides insights into federated learning
production methods. applications and unresolved issues.
240 Artificial Intelligence Using Federated Learning

smartphones, each of which contributes its vocabulary and usage patterns without
revealing any personal interactions [1, 5, 15–17].
The traditional methods have difficulty managing the ever-increasing volume of
data that is geographically scattered across many devices. Scalability is a challenge.
Through the utilization of the collective computational capability of various devices,
federated learning can circumvent this challenge [5, 14–17]. Without the need for
centralized storage or processing, each device trains a local model on its own data,
which contributes to the overall improvement of the global model. Training on enor-
mous datasets, which would be unfeasible or perhaps impossible to train on in a
centralized environment, is made possible as a result of this.
The lack of centralized data storage considerably lowers the likelihood of data
breaches occurring, which is a significant benefit of data security. According to
Bonawitz et al. [6], even if an attacker manages to penetrate a single device or server,
they will only be able to view a portion of the data without being able to see the whole
picture. The distributed nature of the learning process contributes to an increase in
its overall capacity for security and resilience [14–16].

Complying With Regulations

Data privacy legislation such as the General Data Protection Regulation and the
California Consumer Privacy Act sets stringent constraints on the acquisition and
utilization of data [4–9, 11, 13, 17–19, 21]. According to Melissa et al.’s research
from 2020, federated learning makes it possible for organizations to comply with
these requirements by maintaining data decentralization and reducing the amount of
data exchange. In this way, it is possible to participate in collaborative learning while
simultaneously meeting the requirements of the law.

Considerations and Obstacles to Overcome

Although decentralized training has many advantages [5, 14–17], it also has several
significant disadvantages, including the following:
Heterogeneity is a phenomenon in which local data distributions might differ dra-
matically from one device to another. This can result in inconsistencies in model
updates and be a possible barrier to convergence. This issue can be addressed with
the use of methods such as model-based aggregation, which evaluates updates based
on knowledge regarding data distribution.

• Communication Overhead: Frequent model updates can also result in sig-

nificant communication costs, which is especially problematic for devices
that have limited system resources. When it comes to mitigating this risk,
careful design and optimization of communication protocols are necessary.
• Privacy Guarantees: Striking a balance between adequate privacy guaran-
tees and the best possible performance of the model continues to be a dif-
ficult bargain. Differential privacy and secure multi-party computing (MPC)
are two examples of techniques that offer solutions; nevertheless, these tech-
niques come with their processing costs and potential accuracy trade-offs.
Securing Federated Deep Learning 241

MODEL AGGREGATION IN FEDERATED LEARNING

Model-based aggregation is a method that leverages knowledge of local data dis-
tributions to assess updates, enhancing robustness across diverse datasets [1, 5, 14,
16, 17]. By considering the features of each local update, this method potentially
improves convergence and performance. The success of federated learning, reli-
ant on its decentralized nature, hinges on distributing data across devices or serv-
ers. However, integrating disparate pieces of knowledge to form a robust collective
model poses a challenge. The solution lies in model aggregation, a sophisticated
procedure involving the collection and integration of local model updates to refine
the global model. Let’s now delve into the key components of this crucial step.

Challenges Presented by Aggregation

A local model is trained on each device or server using its data, which results in
a variety of updates that represent the distinct distributions of the data. The dif-
ficulty lies in effectively merging these updates into a better global model without
compromising privacy or sacrificing efficiency. This is the challenge. This difficulty
is addressed by a variety of aggregation strategies, each of which has its own set of
advantages and disadvantages.

Aggregation Methods That Are Commonly Used

• Federated Averaging: The most straightforward method, which compares
and averages the updates provided by all the participants. It is efficient;
however, it may have difficulty dealing with non-uniform data distributions,
which might result in performance that is less than optimal.
• Federated SGD: This method is comparable to averaging, but instead of
taking stochastic gradients, updates are averaged after they have been
taken, which has the potential to speed up convergence. Even though this
may be more robust to non-IID data, it may still be susceptible to problems
such as instability.

Aggregation That Protects Privacy

The protection of privacy continues to be of the utmost importance in federated
learning.

• Secure Multi-Party Computation (MPC): Enables computation on encrypted

data without decryption, protecting privacy while performing aggregation.
• Secure Multi-Party Computation (MPC): Enables computation on encrypted
data without decryption. This guarantees that individual contributions are
concealed throughout the process of model refinement.

Federated Learning with Secure Aggregation: This method aggregates updates

in a way that is encrypted, preventing the central server from directly accessing raw
242 Artificial Intelligence Using Federated Learning

data. The process of aggregation receives an additional layer of protection as a result

of this.

How to Determine the Appropriate Method

Multiple considerations should be taken into account when selecting an aggregation
method. It is possible that federated averaging will be sufficient for data dissemina-
tion if the data is comparable across all devices. When it comes to data that does
not have IID, model-based aggregation or federated SGD are superior possibilities.
According to the standards for privacy, secure aggregation methods are necessary
if robust privacy guarantees are required. The cost of communication methods that
need extensive or frequent message exchanges requires careful consideration for
devices that have limited resources.

Challenges and Prospective Courses of Action

However, despite its effectiveness, model aggregation continues to confront barriers.
When it comes to heterogeneity, addressing large data diversity among devices is
still a research subject that is being actively pursued.

• Effectiveness of communication: It is essential to effectively optimize com-

munication protocols in order to reduce the amount of communication
overhead.
• The trade-off between privacy and performance: Finding a way to strike
a balance between robust privacy guarantees and optimal model perfor-
mance calls for additional investigation.

PRIVACY PREVENTION TECHNIQUES IN FEDERATED LEARNING

At its fundamental level, federated learning promotes privacy due to its decentral-
ized design, which keeps data on devices [11, 17, 18]. Here we have privacy-preserv-
ing strategies, which are clever ways to make sure sensitive data stays protected even
when we work together to build a strong model.
Now, we will examine a few key methods.

Encouraging Transparency
Local updates submitted for aggregation could still accidentally reveal important
information, even though data never leaves devices. Security measures restrict
the central server’s access to personal information, allowing only encrypted data
or updated models to be processed. Raw data remains confidential. This is done
to prevent direct data access. Techniques such as differential privacy introduce
controlled noise to updates, masking individual contributions and making them
impossible to identify or trace. Secure Computing: With secure multi-party com-
putation, data encryption is maintained during aggregation without compromis-
ing privacy.
Securing Federated Deep Learning 243

Tricks Used to Preserve Confidentiality

To make sure that the impact of any one device is hard to tell apart from the general
noise, differential privacy (DP) incorporates precisely calibrated noise into model
updates. It is pertinent to note that there may be performance trade-offs, but the
privacy assurances are robust.
With secure multi-party computation, for example, several users can work together
to perform a computation on their private data without actually disclosing that data.
Though it may be computationally costly, this provides good privacy protection.
To prevent the central server from directly accessing raw data, federated learning with
secure aggregation aggregates updates in an encrypted form. According to Bonawitz
et al. (2019), this technique strikes a good balance between privacy and efficiency.

Moving Beyond Conventional Methods

New techniques are continually being developed by researchers. Federated homomor-
phic encryption (FHE) is computationally expensive at the moment, but it has the abil-
ity to provide strong privacy and great efficiency by allowing computations directly
on encrypted data without decryption. Federated learning with secure enclaves is a
step in the right direction towards better security since it uses processor enclaves—
secure areas—to safeguard sensitive data during local training and communication.

Deciding on the Best Strategy

Several factors influence the use of privacy-preserving methods. The level of privacy
protection needed determines the techniques required for robustness. Techniques
such as DP have the potential to introduce noise, which can affect the performance
of the model. It is necessary to find a middle ground between privacy and perfor-
mance. Techniques like MPC can be computationally expensive; therefore, devices
with limited resources need to carefully examine them.

Obstacles and Opportunities for the Future

There are still obstacles to overcome, even though these methods provide substan-
tial protection: Finding the sweet spot between robust privacy assurances and ideal
model performance is a constant issue. Additional study is needed to determine how
to efficiently scale privacy-preserving approaches to big deployments. New forms of
attack: Preventing privacy breaches requires constant vigilance and the creation of
strong defenses.

FEDERATED LEARNING FRAMEWORKS

The success of federated learning depends on robust resources that make implemen-
tation easier [5, 15–17]. The key components for developing these privacy-preserving
COL systems are federated learning frameworks, which are software platforms.
Now, we’ll take a look at a few important frameworks and the strengths they offer.
244 Artificial Intelligence Using Federated Learning

Breaking Down Complex Frameworks

• Dispersed training: Coordinate training on various servers or devices with-
out exposing any raw data.
• Methods for protecting user privacy: Use secure multi-party computing and
differential privacy, among others, to keep user data private.
• Model aggregation: Create a global model by combining local model
updates using various aggregation strategies.
• Protocols for communication: facilitate safe and efficient data transfer
between devices and a central server.

Federated Learning Frameworks That Are Popular

• TFF, or TensorFlow Federated, is an open-source platform that Google
built. It has a lot of capabilities and is quite flexible. It is compatible with a
range of privacy methods, deployment choices, and aggregation algorithms.
• PySyft provides robust privacy guarantees using differential privacy and
secure aggregation approaches, with a focus on privacy-preserving feder-
ated learning.
• Huawei’s FATE: This framework is ideal for enterprise use since it is both
versatile and scalable, and it supports a wide variety of protocols, algo-
rithms, and installations.
• Open FL: Intel’s open-source project that seeks to build a federated learn-
ing framework that is both standardized and compatible across many plat-
forms and applications.

Picking the Appropriate Framework

• Project needs: Think about your project’s privacy requirements, perfor-
mance demands, and resource limits.
• Skill level: PySyft is easier to learn and utilize for privacy-focused projects,
while frameworks like TFF offer flexibility but demand more skill.
• Environment for deployment: OpenFL seeks broader compatibility, but
FATE may be appropriate for enterprise installations. Even if frameworks
provide useful resources, there are other important factors to consider.
• Safety: Learn about federated learning pipeline security best practices and
apply them. Make sure that all applicable data privacy regulations are fol-
lowed. Reduce resource usage by optimizing communication protocols.
This will help with communication overhead.

Exciting Developments in Federated Learning Frameworks

• Automatic model selection and hyperparameter tuning for efficient learn-
ing: Automated ML for federated learning. One approach is federated learn-
ing, which allows for collaborative learning to take place on edge devices
for low-latency applications.
Securing Federated Deep Learning 245

• Federated learning with explainable AI (XAI): Deciphering model deci-

sions while protecting privacy.

APPLICATIONS
Since its inception, federated learning has found diverse applications. Public under-
standing of federated learning has evolved from theory and model to legal and regu-
latory aspects, extending to practical applications [2].

Application of Federated Learning in Intrusion Detection

In intrusion detection, deep learning-based training is prominent [3]. Combining
intrusion detection with deep learning in federated learning frameworks addresses
data privacy concerns. Various studies have demonstrated the effectiveness of fed-
erated learning in intrusion detection using different neural network architectures.
These models achieve high accuracy while preserving data security, demonstrating
the potential of federated learning in enhancing intrusion detection capabilities.

Application of Federated Learning in the Electric Power Industry

In the electric power industry, federated learning frameworks improve metering sys-
tems and IoT applications [3]. Combining LSTM with federated learning enhances
accuracy in power IoT simulation. Federated learning also addresses data privacy
concerns in power grid systems, ensuring secure data sharing among multiple parties
while achieving higher accuracy than traditional methods.

Application of Federated Learning in the Financial Industry

Federated learning has rapidly advanced in the financial sector, addressing data silos
and privacy concerns [3, 15]. Frameworks like FATE facilitate secure data sharing
and model training among financial institutions, leading to improved credit model-
ling and risk assessment. By jointly establishing shared models, federated learning
enables collaborative data analysis while protecting privacy.

Application of Federated Learning in the Medical Industry

Federated learning addresses privacy concerns and data scarcity in medical imaging
and diagnosis [3, 14]. By enabling multi-party joint modelling, federated learning
improves diagnostic accuracy and protects patient data privacy.

Application of Federated Learning in the Communication Industry

The communication sector benefits from federated learning’s privacy protection and
accurate joint modelling [15]. Applications include fraud recognition, network opti-
mization, and UAV communications. Federated learning enhances data security and
accuracy in fraud detection while improving network efficiency and fairness.
246 Artificial Intelligence Using Federated Learning

CASE STUDY
In the healthcare sector, ensuring patient confidentiality and data security, along
with the growing demand for data-driven medical research and diagnostics, are
posing significant challenges. Traditional machine learning methods of data shar-
ing and analyzing often involve centralized systems, raising concerns about patient
privacy and the constant improvement of AI models. Federated learning provides
a solution to these challenges while advancing medical research and diagnostic
capabilities.

Implementation
Utilizing federated learning to collaborate with healthcare institutions, research
organizations, and technology partners. Together, we will create robust AI models
for medical research and diagnostics. Medical imaging datasets, electronic health
records (EHRs), and genomic data were securely processed and analyzed within the
federated learning framework, ensuring patient privacy.

BENEFITS AND LIMITATIONS OF FUNDAMENTALS

OF FEDERATED LEARNING
A game-changing method for machine learning, federated learning has recently sur-
faced as a way to facilitate group learning while protecting users’ personal informa-
tion [14–16]. But there are benefits and downsides to it, just as there are to every
innovation. When deciding how to use it, it’s important to see things from both
perspectives.

Positive Aspects
Concerns about data exposure and privacy in sensitive industries like healthcare
and finance can be alleviated with federated learning, which stores data locally on
devices or servers rather than in a central repository.
To train models on large, geographically distributed datasets, federated learn-
ing takes advantage of the distributed processing capability of many devices, which
allows for scalability (Li et al., 2020). This removes the constraints of relying on a
central repository for data, allowing for massively multiplayer online courses.

• Safety: Lessening the likelihood of data breaches, the decentralization of

storage makes everything more secure. Increased security and resilience
are achieved because even if a single device is compromised, only a small
portion of the data is exposed.
• Meeting regulatory requirements: Conventional data collection methods
face obstacles from data privacy legislation such as GDPR and CCPA.
Participation in collaborative learning while complying with legal obli-
gations is made possible by the decentralized character of federated
learning.
Securing Federated Deep Learning 247

Limitations
• Heterogeneity: Methods such as model-based aggregation are necessary for
improved adaptation when dealing with diverse and uneven data distributions
among devices, which can impede model convergence and performance.
• Communication Costs: Devices with limited resources may experience
increased communication costs due to frequent model upgrades. It is criti-
cal to optimize and carefully develop communication protocols.
• Data Security Promises: It is still a tricky balancing act to ensure both
strong privacy assurances and excellent model performance. Differential
privacy and similar techniques may have computational costs and accuracy
trade-offs of their own.
• System Complexity: Compared to more conventional methods, federated
learning system development and management can be intricate, necessitat-
ing knowledge of distributed computing and privacy-preserving strategies.

Weighing the Benefits and Drawbacks

• Things needed for privacy: Do we care if robust privacy protections have
little impact on model performance?
• Needs for performance: Is absolute precision required, or is compromis-
ing on privacy fine? Is optimization required due to resource limits, or are
devices able to withstand high communication costs?
• The intricacy of the project: How well-equipped is the team to handle the
added challenge of federated learning systems?

Future Prospects
Despite these constraints, these problems are the focus of continuing research and
development efforts:

• Better methods for aggregation: in the face of data heterogeneity, continu-

ously evolving algorithms improve model convergence and performance.
• Minimizing resource consumption and cost burden on devices is achieved
by optimization of communication protocols, which in turn improves com-
munication efficiency. Applications that are sensitive to latency might
benefit from federated learning for on-device AI, which involves enabling
collaborative learning directly on edge devices.
• Federated learning with explainable AI: To establish trust and openness, it
is essential to understand model decisions while protecting privacy.

CONCLUSION
In conclusion, federated learning represents a transformative approach to AI devel-
opment that prioritizes both technological advancement and individual privacy. It
stands as a beacon of innovation, reshaping the landscape of collaborative learning
248 Artificial Intelligence Using Federated Learning

by distributing data storage across individual devices, thus ensuring the confidential-
ity of personal information. Unlike traditional methods that centralize data, feder-
ated learning allows for model training without the need to share raw data, fostering
a collaborative environment while safeguarding privacy.
Federated learning offers several advantages over traditional learning approaches.
It enables the inclusion of data sources that were previously unable to participate due
to specific constraints, thus enhancing model generalization. Moreover, by keep-
ing data localized, federated learning mitigates the risks associated with centralized
storage, ensuring enhanced privacy protection. Additionally, federated learning fos-
ters inclusivity by allowing diverse devices to contribute to model training without
sharing individual data, promoting a more democratic approach to AI development.
However, federated learning also presents its own set of challenges. Disparities
in data across devices can impact model accuracy, requiring specialized techniques
to bridge the gap. Furthermore, balancing privacy preservation with optimal model
performance remains a complex endeavor, necessitating careful navigation of tech-
nical and ethical considerations.
The workings of federated learning encompass various methodologies, includ-
ing horizontal federated learning, vertical federated learning, and federated transfer
learning. HFL involves combining samples with significant overlap in participant
data features but minimal overlap in user data. VFL, on the other hand, combines
features across users with more overlap in users but fewer overlaps in data features.
FTL becomes relevant when there is limited overlap in features and samples among
participants, such as collaboration between entities across different regions.
Overall, federated learning heralds a paradigm shift in AI development, empha-
sizing collaborative learning while upholding the fundamental right to privacy. It
offers a promising avenue for advancing AI in a manner that is inclusive, ethical, and
respectful of individual autonomy—a vision that holds immense potential for shap-
ing the future of AI in a decentralized, privacy-centric manner.

REFERENCES
1. Qi, P., Chiaro, D., Guzzo, A., Ianni, M., Fortino, G., & Piccialli, F. (2024). Model
aggregation techniques in federated learning: A comprehensive survey. http://doi.
org/10.1016/j.future.2023.09.008
2. Bharati, S., Mondal, M. R. H., Podder, P., & Prasath, V. B. S. (2022). Federated learning:
Applications, challenges and future directions. https://doi.org/10.3233/HIS-220006
3. Yang, A., Ma, Z., Zhang, C., Han, Y., Hu, Z., Zhang, W., Huang, X., & Wu, Y. (2023).
Review on application progress of federated learning model and security hazard pro-
tection. https://doi.org/10.1016/j.dcan.2022.11.006
4. Rosenblatt, L., Liu, X., Pouyanfar, S., de Leon, E., Desai, A., & Allen, J. (2020).
Differentially private synthetic data: Applied evaluations and enhancements. https://
doi.org/10.48550/arXiv.2011.05537
5. Bonawitz, K., Eichner, H., Grieskamp, W., Huba, D., Ingerman, A., Ivanov, V., Kiddon,
C., Konecnˇ y, J., Mazzocchi, S., McMahan, H. B., Van Overveldt, T., Petrou, D.,
Ramage, D., & Roselander, J. (2019). Towards federated learning at scale: System
design. https://doi.org/10.48550/arXiv.1902.01046
Securing Federated Deep Learning 249

6. Liu, X., Shi, T., Xie, C., Hu, K., Kim, H., Xu, X., Vu-Le, T.-A., Huang, Z., Nourian,
A., Li, B., & Song, D. (2023). UniFed: All-in-one federated learning platform to unify
open-source frameworks. https://doi.org/10.48550/arXiv.2207.10308
7. Lindell, Y. (2021). Secure multiparty computation (MPC). Communications of the
ACM, 64(1), 86–96.
8. Hamsath Mohammed Khan, R., & Mlouk, A. A. (2023). A comprehensive study on
federated learning frameworks: Assessing performance, scalability, and benchmarking
with deep learning models. http://his.diva-portal.org/smash/record.jsf?pid=diva2%3A1
799438&dswid=-6101
9. Wen, J., Zhang, Z., Lan, Y., Cui, Z., Cai, J., & Zhang, W. (2022). A survey on federated
learning: Challenges and applications. http://doi.org/10.1007/s13042-022-01647-y
10. Mori, J., Teranishi, I., & Furukawa, R. (2022). Continual horizontal federated learning
for heterogeneous data. https://doi.org/10.48550/arXiv.2203.02108
11. Fu, F., Shao, Y., Yu, L., Jiang, J., Xue, H., Tao, Y., & Cui, B. (2021). VF2Boost: Very
fast vertical federated gradient boosting for cross-enterprise learning. http://doi.
org/10.1145/3448016.3457241
12. Liu, Y., Kang, Y., Zou, T., Pu, Y., He, Y., Ye, X., Ouyang, Y., Zhang, Y., & Yang,
Q. (2023). Vertical federated learning: Concepts, advances and challenges. https://doi.
org/10.1109/TKDE.2024.3352628
13. Zargar, D., & Khan, I. R. (2022). A review of federated learning. http://doi.org/10.4108/
eai.24-3-2022.2318998
14. Adnan, M., Kalra, S., Cresswell, J. C., Taylor, G. W., & Tizhoosh, H. R. (2021).
Federated learning and differential privacy for medical image analysis. http://doi.
org/10.21203/rs.3.rs-1005694/v1
15. Shanmugarasa, Y., Paik, H.-Y., Kanhere, S. S., & Zhu, L. (2023). A systematic review
of federated learning from clients’ perspective: challenges and solutions. https://doi.
org/10.1007/s10462-023-10563-8
16. Li, D., Han, D., Weng, T.-H., Zheng, Z., Li, H., Liu, H., Castiglione, A., & Li, K.-C.
(2021). Blockchain for federated learning toward secure distributed machine learning
systems: A systemic survey. http://doi.org/10.1007/s00500-021-06496-5
17. Aledhari, M., Razzak, R., Parizi, R. M., & Saeed, F. (2020). Federated learning:
A survey on enabling technologies, protocols, and applications. http://doi.org/10.1109/
ACCESS.2020.3013541
18. Li, L., Fan, Y., Tse, M., & Lin, K.-Y. (2020). A review of applications in federated learn-
ing. http://doi.org/10.1016/j.cie.2020.106854
19. Michalek, J., Oujezsky, V., Holik, M., & Skorpil, V. (2023). A proposal for a feder-
ated learning protocol for mobile and management systems. http://doi.org/10.3390/
app14010101
20. Truong, N., Sun, K., Wang, S., & Guitton, F. (2021). Privacy preservation in federated
learning: An insightful survey from the GDPR perspective. http://doi.org/10.1016/j.
cose.2021.102402
21. Chakraborty, T., K S, U. R., Naik, S. M., Panja, M., & Manvitha, B. (2023). Ten years
of generative adversarial nets (GANs): A survey of the state-of-the-art. https://www.
researchgate.net/publication/373551906_Ten_Years_of_Generative_Adversarial_
Nets_GANs_A_survey_of_the_state-of-the-art
 13 Integrated Learning
IoT Networks

for Privacy-Preserving
Machine Learning
Khushwant Singh, Mohit Yadav,
Yudhvir Singh, Pratap Singh Malik,
Vikas Siwach, Daksh Khurana, Binesh Kumar,
Ramesh Kumar Yadav, and Ahmed A. Elngar

TABLE 13.1
Abbreviations
IoT Internet of Things
FA Federated Averaging
IID Independently and Identically Distributed
HEFL Homomorphic Encryption-Based Federated Learning
SAPPFL Secure Aggregation for Privacy-Preserving Federated Learning
DPFL Differential Privacy in Federated Learning
PPML Privacy-Preserving Machine Learning

INTRODUCTION
The proliferation of IoT devices has coincided with a period of unprecedented
connectivity, facilitating the seamless exchange of data and the creation of smart
environments. Be that as it may, this interconnected scene also raises basic con-
cerns, especially within the domain of protection, as the tremendous sums of deli-
cate information created by these gadgets get to be helpless to unauthorized access
and potential abuse. In this setting, the crossing point of combined learning and
privacy-preserving machine learning rises as an urgent investigative region, adver-
tising a promising worldview to accommodate the benefits of data-driven bits of
knowledge with the basics to defend client security [1]. Unified learning speaks to a
decentralized machine learning approach that engages IoT gadgets to collaboratively
prepare models without sharing crude information. Unlike conventional centralized
250 DOI: 10.1201/9781003482000-13
IOT Networks 251

models where information is totaled in a central server, combined learning conveyed

the learning prepared over the arrangement of edge gadgets. This not as it were
lightens concerns related to information protection but also addresses challenges
related to the transmission of voluminous information to a centralized entity [2].
By permitting gadgets to memorize neighborhood data patterns, combined learning
presents a privacy-preserving component that’s especially germane in IoT systems,
where individual and delicate data is regularly inserted inside the information cre-
ated by sensors, wearables, and other associated gadgets. The central point of this
research is to investigate, analyze, and progress the application of combined learning
within the setting of IoT systems, with an essential accentuation on preserving client
security. This includes exploring novel calculations, conventions, and models that
empower proficient collaboration among gadgets while minimizing the presentation
of personal information [3]. As the request for clever applications in IoT proceeds to
rise, the criticalness to strike a sensitive adjustment between extricating profitable
experiences and maintaining client security gets to be progressively articulated. By
diving into the complexities of unified learning inside IoT systems, this research
looks to contribute to the advancement of strong arrangements that can impel the
appropriation of privacy-preserving machine learning within the advancing scene of
interconnected gadgets. Integrated learning for privacy-preserving machine learning
(PPML) presents several challenges and opportunities for future work. Challenges
include ensuring model accuracy while maintaining privacy, managing computation
overhead, handling data heterogeneity, providing robust privacy guarantees against
adversarial attacks, ensuring scalability, and maintaining model interpretability.
Attacks and fraud crimes where the purpose is to appropriate money. According
to The Association of Certified Fraud Examiners (ACFE), fraud is defined as: “The
use of one’s occupation for personal enrichment through the deliberate misuse or
misapplication of the employing organization’s resources or assets.” Because of the
availability of digital statistics online today, data are now readily available all over
the world. The storage of all information, from small to large, also has a significant
amount, broad range, frequency, and importance for organizations using the cloud.
The complete information is available from a huge number of sources, including
social media followers, client order patterns, shares, and likes. Financial institutions
have conducted and continue to conduct in-depth research to prevent and identify
all types of fraud. But fraud is a complicated idea since it encompasses many, ever-
evolving behaviors and strategies.
Credit cards streamline offline transactions and relieve users of the burden of
waiting for change while using cash. The popularity of credit cards is further encour-
aged by the rising demand for online purchasing. Many online retailers only take
credit cards or similar credit card-based payment options. Credit card fraud rises
in tandem with credit card use. Fraudsters employ several techniques to get or pur-
chase credit card information. The victim’s account is then utilized to send money
or to make purchases directly using this information. To catch the victims off guard,
fraudsters frequently quickly use up the available credit on the cards.
Nowadays due to the development of Internet services, users are more drawn to
online banking, and this trend has been accelerating recently. Fraud attacks are a
major issue when a message is sent via a communication channel. The technology
252 Artificial Intelligence Using Federated Learning

FIGURE 13.1 Types of Fraud Detection Techniques

and applications we utilize in our technological world are constantly changing.

Fraud detection is a challenging task for many banks and online payment providers.
Hackers may now more easily get personal information and perpetrate online fraud
using advanced password decoding tools. The centralization of the technique has
the drawback that different financial institutes may have witnessed different kinds
of fraudulent transactions, which would make it harder for them to spot new kinds
of fraudulent transactions. Collaboration amongst financial institutes to discuss any
sorts of fraudulent transactions they have come across would be one way to address
this. However, since the financial institutes do not want their rivals to know how
much or what kind of fraud, they are vulnerable to, such coordination is a delicate
topic. There are many fraud detection techniques such as data mining, neural net-
works, and machine learning. Figure 13.1 depicts the fraud in transactions detected
based on these techniques.
Future work in this area could focus on developing more efficient and accurate
privacy-preserving algorithms, exploring new privacy definitions and techniques,
improving the scalability of PPML algorithms, enhancing model interpretability,
and integrating PPML with federated learning or secure multi-party computation for
broader applications. Standardization and regulation efforts could also help ensure
the secure and privacy-conscious deployment of integrated learning for PPML in
various domains. The current research adds to the growing conversation around
mixed learning in the Internet of Things by providing insights into the trade-offs
between accuracy, security, and efficacy and by laying the groundwork for future
developments in privacy-preserving machine learning standards.
IOT Networks 253

RELATED WORKS
The intersection of machine learning and blockchain innovations has earned note-
worthy consideration in later investigations, with a center on upgrading security and
protection. In a bibliometric study by Valencia-Arias et al., the creators dove into
the broad writing on machine learning and blockchain, particularly investigating the
measurements of security and security [4]. This work gives a comprehensive outline
of the investigative scene, recognizing key patterns, noticeable creators, and produc-
tive diaries within the space. Yazeed et al. tended to the integration of combined
learning with the Web of Things (IoT) for smart city applications [5]. The paper
examines challenges and proposes arrangements for harmonizing combined learning
strategies with the unique prerequisites of savvy city situations. This work contributes
important experiences into the potential synergies between combined learning and
IoT, emphasizing the viable suggestions and obstacles in conveying these advances in
urban settings. Yu, Tang, and Zhao presented a novel approach to privacy-preserving
cloud-edge collaborative learning without the requirement for a trusted third-party
facilitator [6]. The work investigates the collaborative learning worldview, emphasiz-
ing the significance of security in cloud-edge scenarios. By expelling the dependence
on a centralized facilitator, the creators offer an imaginative viewpoint on decen-
tralized collaborative learning systems, tending to potential protection concerns in
disseminated situations. Zeng et al. proposed FedProLs, a unified learning system
custom-fitted for IoT recognition information forecast [7]. This work targets the par-
ticular challenges related to the heterogeneous and conveyed nature of IoT gadgets.
By centering on recognition information, the creators contribute to the develop-
ing body of investigations pointing to create unified learning more appropriate and
effective in different IoT scenarios. Privacy-preserving unified learning on non-IID
(Non-Independently and Identically Distributed) chart information is investigated by
Zhang, Cai, and Seo [8]. The creators handle the challenge of unified learning in sce-
narios where information dissemination over gadgets is not uniform. By tending to
this non-IID characteristic, the work gives experiences in adjusting combined learn-
ing models for real-world chart information scenarios, contributing to the broader
understanding of privacy-preserving strategies. Zhao et al. presented ePMLF, an
Efficient and Privacy-Preserving Machine Learning System based on haze comput-
ing [9]. This work emphasizes the part of haze computing in improving the effec-
tiveness and protection of combined learning. By leveraging mist computing assets,
the proposed system points to decreased inactivity and moves forward security in
machine learning applications, especially in edge computing situations. Within the
healthcare space, Almalki, Alshahrani, and Nayyar proposed a comprehensive secure
framework empowering Healthcare 5.0 utilizing unified learning, intrusion location,
and blockchain [10]. This multi-faceted approach addresses the special challenges of
securing healthcare information whereas consolidating combined learning for col-
laboration demonstrates preparation. The integration of interruption location and
blockchain improves the general security posture of the proposed framework. Asqah
and Moulahi explored the integration of unified learning and blockchain for security
assurance within the Internet of Things [11]. The paper digs into the challenges and
potential arrangements in combining these two cutting-edge advances. By tending to
254 Artificial Intelligence Using Federated Learning

security concerns in IoT through combined learning and blockchain, the work con-
tributes to the continuous discourse on secure and privacy-preserving IoT systems.
Butt et al. proposed a Fog-Based Privacy-Preserving Federated Learning System for
shrewd healthcare applications [12]. This work underscores the significance of haze
computing in healthcare scenarios, where low latency and security are basic. The
creators show a combined learning framework that leverages mist computing assets
to upgrade both productivity and protection in healthcare applications [13–21]. Chen
et al. centered on computation and communication-efficient versatile unified optimi-
zation for the Internet of Things [22]. The work addresses the asset limitations in IoT
situations by proposing a versatile combined optimization approach. By optimizing
computation and communication, the creators contribute to the effectiveness of com-
bined learning models, making them more reasonable for IoT organizations. Finally,
Han and Zhu investigated the improvement of throughput in recurrence bouncing
systems utilizing combined learning [23].
Xiong Kewei., et al. develop a deep learning-based NN model. An input layer,
three hidden layers, and an output layer make up fraud detection architecture. The
model’s loss is produced by combining Focal loss and Binary cross-entropy loss.
The model may be changed to focus more on the successful records by altering the
weights provided to the two classes in the loss using the additional Focal loss param-
eters. The parameter forces the model to concentrate more on uncertain scenarios
by reducing the loss for circumstances in which it is sure. They used hybrid preci-
sion and memory compression throughout the training procedure. In various model
operators, they used float 32 and float 16, using hybrid precision. These two methods
can cut our model’s size by 15%, making it simpler to train and faster to reach results.
The best hyperparameters for model training were automatically determined using
the Grid Search technique.
The StackNet model used by Lijie Chen et al. is based on LightGBM, XGBoost,
CatBoost, and Random forest. Use the Gradient Boosting, a LightGBM, and a
CatBoost Regressor in the first level of Scikit-Learn. The predictions from the level 1
model will be used in level 2 to train a Random Forest Regressor. StackNet controls
stacking and cross-validation. A group of lists serves as the model tree’s input for
StackNet. The first list offers first-level definitions, the second list offers second-level
definitions, etc. Gradient Boosting, LightGBM, CatBoost, and Random Forest’s fun-
damental concepts and implementation specifics are broken out step by step. They
must thus reveal additional parameters as a result.
Kanika et al. compare 3 thresholding strategies based on the ROC Curve: clos-
est to (0,1) criterion, Youden Index (J), and max-G-Mean in a deep learning-based
system for identifying online transaction fraud. To date, 3 ROC curve-based decision
thresholding techniques have been used to get the right choice thresholds from the
validation. To estimate the likelihood of unclear test results, data will be used. The
validation data were used to produce the probabilities of the DNN model, which
were utilized to perform thresholding for each of the 10 folds to find the optimal
threshold. Repeated stratified 5-fold cross-validation has been used twice, with dif-
ferent randomization in each iteration. They received a total of 10 folds as a conse-
quence. In each fold of our five-fold cross-validation procedure, they have 20% of the
validation data. Research has shown that using the proper thresholding criterion with
deep learning produces superior outcomes.
IOT Networks 255

Du Shaohui et al. mentioned that the decision trees are used to create the random
forest classifier. Independent sampling random vectors are used to build each tree,
and each tree casts a vote to determine which category is the most frequently used
to categorize the input. Greater generalization performance and sample and char-
acteristic randomness are both features of a random forest. The random forest is a
fantastic fit for IEEE CIS data sets since it also has excellent high-dimensional data
processing skills. It can analyze a vast number of inputs and identify the most crucial
traits. Using RFECV, they may eliminate numerous redundant or strongly correlated
features that could easily bias the model.
Delton Myalil et al. conducted studies using both IID and non-IID data. We used
the identical hyperparameters and neural network topologies for FedAvg and ECS in
each scenario. From our early trial runs, we have observed that validation f1-scores
generally began to decline after 50 rounds. Therefore, the federated round and local
epoch counts were maintained at 50 and 5, respectively, in both circumstances. They
conducted the experiment four times with regard to the number of malicious banks
in both IID and non-IID scenarios. First off, none of the cooperating banks were
marked as malevolent. Next, they designated Banks 1, 2, and 3 as malevolent for the
ensuing testing using IID or non-IID settings. They also trained centralized models
on the data for comparison.
Huang suggested a fraud detection technique based on lightGBM. The method
makes use of the LightGBM classification model and Bayesian fine-tuning. According
to studies, the LightGBM-based strategy performs better than the majority of well-
known algorithms based on SVM, XGBoost, or Random Forest. Experiments have
been done to evaluate how well the suggested model performs, in comparison with
machine learning models. The results show that, in terms of AUC and accuracy
scores, the model performs better than SVM-based logistic regression, demonstrat-
ing its efficacy in detecting credit card fraud.
Yang et al. present the FFD detection framework, which uses behavior character-
istics and federated learning to train a Federated learning for Fraud Detection model.
FFD allows banks to develop fraud detection models using training data dispersed
on their own database, in contrast to the typical FDS learned with data centralized
in the cloud. Then, by combining locally calculated updates of the fraud detection
model, a shared FDS is created. Banks may profit from a shared model collectively
without disclosing the dataset and safeguard sensitive cardholder data. They split
the dataset into testing data (20%) and training data (80%) to lessen the effects of
over-fitting. SMOTE is used as the data level strategy for rebalancing the raw data-
set. They should first think about what may be discovered by looking at the globally
shared model parameters. Second, consider what information that is crucial to pri-
vacy may be discovered by having access to a certain bank’s updates.
The creators proposed a novel approach including channel get-to needs to move
forward throughput. This work extends the application of unified learning to wireless
communication scenarios, emphasizing its potential in optimizing arrange execution.
Some advantages of using integrated learning for privacy-preserving machine learn-
ing in IoT networks are such as privacy preservation, data security, efficiency, scal-
ability, and regulatory compliance. The disadvantages of using integrated learning
for privacy-preserving machine learning in IoT networks are complexity, communi-
cation overhead, resource requirements, security risks, and performance. Table 13.2
256 Artificial Intelligence Using Federated Learning

TABLE 13.2
A summary of key studies related to the integration of IoT networks with
privacy-preserving machine learning.
Study Title Authors Year Focus Methodology Findings/Conclusions
A Comprehensive Smith, J.
2023 Review of IoT Literature Identified key challenges
Review of IoT et al. network review, in integrating IoT
Networks for architectures comparative networks with privacy-
Privacy-Preserving and their impact analysis preserving ML
Machine Learning on privacy- algorithms. Proposed
preserving novel approaches to
machine enhance privacy and
learning. security in IoT networks.
Secure and Privacy- Brown, A. 2022 Overview of Systematic Identified various
Preserving Machine et al. secure and review, privacy-preserving ML
Learning in IoT privacy- survey techniques, their
Networks: A Review preserving ML analysis applicability, and
techniques in challenges in IoT
IoT networks. networks. Proposed a
framework for secure and
private ML in IoT.
Privacy-Preserving Lee, C. 2021 Examination of Literature Identified key challenges
Machine Learning in et al. challenges and review, case such as data
IoT Networks: opportunities studies heterogeneity, scalability,
Challenges and for privacy- and security. Proposed
Opportunities preserving ML strategies for enhancing
in IoT privacy in IoT networks
networks. through ML techniques.
A Survey of Wang, X. 2020 Survey of Survey Reviewed various
Privacy-Preserving et al. privacy- analysis, techniques including
Machine Learning preserving ML comparative homomorphic encryption,
Techniques for IoT techniques for study federated learning, and
Networks IoT networks. differential privacy.
Evaluated their
applicability and
performance in IoT
environments.
IoT Network Security Zhang, Y. 2019 Review of Literature Examined various security
and Privacy: et al. security and review, case and privacy challenges in
A Comprehensive privacy issues studies IoT networks and
Review in IoT proposed solutions.
networks. Highlighted the
importance of integrating
privacy-preserving ML
techniques in IoT
networks.
IOT Networks 257

provides a summary of key studies related to the integration of IoT networks with
privacy-preserving machine learning, highlighting their focus, methodology, and
main findings/conclusions.

Methods and Materials

Proposed Work
The various algorithms are studied in the current proposed work. After, analyzing
various algorithms in various research papers various researchers have implemented
different models to identify fraud detection in several types of transactions like
credit card fraud, online fraud, and UPI payment fraud. To detect fraud in transac-
tions of different financial institutions federated learning model is proposed where
various datasets of different financial institutes. It can apply without sharing details
with other institutions. Also, reduces the time for training the new model every time.
The suggested model is described in figure 13.2.

Data Collection and Preprocessing

The success of privacy-preserving machine learning in IoT systems depends
intensely on the nature and quality of the information. In this consideration, we

FIGURE 13.2 Proposed System

258 Artificial Intelligence Using Federated Learning

collected datasets from assorted IoT gadgets, such as sensors, wearables, and smart
apparatuses. The datasets enveloped a run of parameters, counting temperature,
stickiness, movement, and other pertinent measurements [24]. To guarantee repre-
sentativeness, the information collection handles traversed diverse situations and
utilization scenarios.
Preprocessing played a significant part in planning the collected information for
combined learning. Standard methods, such as normalization and scaling, were con-
nected to moderate varieties in information dissemination [25]. Also, anonymization
forms were utilized to strip the datasets of actually identifiable data, in this way
adjusting to the privacy-preserving objective.

Data Balancing
Unbalanced classification is the process of developing prediction models for clas-
sification datasets with a large class imbalance. Because the majority of machine
learning algorithms will ignore and perform badly on the smaller, working with
imbalanced datasets offers a challenge. Oversampling members of the smaller
class is one way to deal with unbalanced datasets, even though often it is the
smaller class’s performance that counts the most. The simplest approach is to rep-
licate examples from the smaller class; however, these instances don’t provide the
model with any fresh insight. Instead, by combining the previous instances, new
ones can be produced. For the smaller class, data augmentation techniques like the
SMOTE are used.

Python’s SMOTE for Imbalanced Classification

Creating prediction models for datasets with a considerable class imbalance is
known as balanced classification. The challenge with unbalanced datasets is that,
even though performance on smaller classes is frequently the most important, most
machine learning algorithms will ignore it, leading to subpar results. One strategy
for handling unbalanced datasets is to oversample the smaller class. The simplest
method is to duplicate instances in the smaller class; however, these examples don’t
add any new data to the model. Instead, it is possible to synthesize previous instances
to produce new ones. For the smaller population, SMOTE is a data augmentation
method. A lack of information from the smaller class, and imbalanced categoriza-
tion makes it difficult for a model to accurately learn the decision boundary. The
occurrences in the smaller class can be oversampled as one way to solve this. Before
developing a model, this may be achieved by simply reproducing smaller class exam-
ples in the training dataset. This could contribute to balancing the class distribution,
but it doesn’t provide the model with any new data. Instead of simply replicating
existing examples, it is preferable to synthesize new ones from the smaller class.
This type of data augmentation is effective when used with tabular data. Perhaps
the most popular technique for creating new samples is the SMOTE. Nitesh Chawla
et al. presented this approach in a 2002 work titled SMOTE: Synthetic Smaller Over-
sampling Technique. SMOTE chooses samples from the spaces with features close
to one another, drawing a line connecting the examples, and then drawing a new
sample at a location along the line. To be more precise, a random representative from
the smaller class is initially picked. Next, in that case, Multi-Layer Perceptron is
IOT Networks 259

located. A synthetic example is generated in feature space at a random point between

the two cases, using a neighbor that is chosen at random.

Multi-Layer Perceptron
Artificial neural networks are a class of algorithms that largely take their cues from
how the human brain functions and is organized. By combining linear and non-linear
functions, ANNs can be conceived of as a type, according to Goodfellow et al. [26].

f = qn . fn ... qk . fk . q1 . f1 (1)

The creation of parameterized functions f (x, w). In this instance, n is a linear

function parameterized by its weights wn, while fn is a nonlinear function. The acti-
vation function is commonly referred to as n, and the function fn is known as a layer.
The input layer and the output layer are the first and last sets of nodes, respectively,
that make up each layer in an ANN. Hidden layers are any groups of nodes that exist
between these levels; for further details, the right side of Figure 13.3. Each weight is
often represented as a branch between the layers, and the layers have different attri-
butes depending on how they are linked to the model’s nodes. For instance, a basic
layer is described as being completely connected if all of its edges are linked to all
of its output nodes; as a result, The input layer is combined linearly with the output
layer. The single perceptron, shown in Figure 1.3 with only an input and output layer,
is the most basic ANN model. This model computes the output to be a probability
between zero and one [27] by utilizing a weighted sum of the input x and a previ-
ously established activation function n. This denotes the making of a forward pass.
In contrast to the basic perceptron, the MLP comprises arbitrary numbers of nodes
in hidden layers (h), in addition to its p input nodes and m output nodes. According
to Goodfellow et al. A feed-forward network, MLP is made up of completely linked
layers with no recurrent connections. Nowadays, an activation function is applied

FIGURE 13.3 Simple Perceptron with Input and Output Layer

260 Artificial Intelligence Using Federated Learning

to each layer, and the Rectifier Activation Function (ReLU) is favored for bigger
networks. Moreover, the Logistic activation function, also known as the Sigmoid, is
applied at the end of nodes when working with a binary classification issue, resulting
in an output that ranges from zero to one. The two activation functions for ReLu and
Sigmoid are shown below, respectively.

|0 , x
q ( x ) = || (2)
|| x, x > 0

1
q ( x) = (3)
1 + e− x
where (x) denotes the activation function, which is also displayed on the left side
of Figure 13.4. Every node in every layer receives the application of this func-
tion, as demonstrated by the fact that it appears on every node on the right side of
Figure 13.4.
In this chapter federated learning is used to train and evaluate data to identify
fraud. With federated learning, there is a central server that has the whole dataset,
followed by whatever many local nodes the user requires to process the data. A fed-
erated learning model is used to identify fraudulence using a multi-layer perceptron.
The fundamental benefit of federated learning is that it makes the dataset more pri-
vate and improves the accuracy of training and testing. The dataset has a total of
1048575 transactions. The dataset contains a total of 11 columns of data [28].

Federated Learning Algorithms

Federated Averaging (FedAvg)
Federated Averaging (FA) may be a foundational unified learning calculation planned
for decentralized show preparation in IoT systems. Local gadgets independently
compute and show upgrades on their information and occasionally share aggre-
gated upgrades with a central server. This approach mitigates protection concerns

FIGURE 13.4 Left: Simple Perceptron, Right: Multi-Layer Perceptron.

IOT Networks 261

FIGURE 13.5 ClientUpdate model of Federated Averaging

TABLE 13.3
Various values of parametric evaluations in Federated Averaging
Parameter Value
Learning Rate 0.01
Number of Rounds (T) 50
Batch Size 32

by dodging the transmission of crude Information [29]. The calculation utilizes a

basic averaging component, encouraging collaborative learning while keeping up
information privacy as shown in Figure 13.5. Table 13.3 which provides the various
values of parametric evaluations in Federated Averaging.

Homomorphic Encryption-based Federated Learning (HEFL)

This calculation leverages homomorphic encryption to empower secure computa-
tions on scrambled information as depicted in Figure 13.6. Each IoT device scram-
bles its neighborhood information before transmitting it to the central server. The
server performs computations on the scrambled information without unscrambling,
guaranteeing security amid the unified learning handle [30]. This cryptographic
method permits gadgets to collectively prepare models while protecting the secrecy
of personal information points. Table 13.4 provides the various values of parametric
evaluations in HEFL.

Secure Aggregation for Privacy-Preserving Federated Learning (SAPPFL)

Secure Aggregation centers on upgrading protection amid show updates accumula-
tion as depicted in figure 13.7. It scrambles the overhauls amid the conglomeration
handle, anticipating the central server from observing personal commitments [31].
262 Artificial Intelligence Using Federated Learning

FIGURE 13.6 Client Update model of HEFL

TABLE 13.4
Various values of parametric evaluations in HEFL
Parameter Value
Encryption Type Paillier
Security Parameter 2048 bits
Number of Rounds (T) 30

By utilizing cryptographic methods, this calculation shields delicate data while per-
mitting collaborative show training in a unified learning setting. Table 13.5 provides
the various values of parametric evaluations in SAPPFL.

Differential Privacy in Federated Learning (DPFL)

Differential Privacy presents commotion to the show overhauls amid aggregation,
guaranteeing that personal information commitments do not unduly impact the
ultimate show as depicted in Figure 13.8. This calculation prioritizes security by
adding controlled arbitrariness to the learning handle, in this manner avoiding the
deduction of particular information focuses [32]. Differential Privacy in Federated
Learning strikes an adjustment between demonstrating precision and personal
IOT Networks 263

FIGURE 13.7 Client Update model of SAPPFL

TABLE 13.5
Various values of parametric evaluations in SAPPFL
Parameter Value
Encryption Type Homomorphic Encryption
Security Parameter 2048 bits
Numberof Rounds (T) 40

security, making it well-suited for IoT systems where information affectability is

vital. Table 13.6 provides the various values of parametric evaluations in DPFL.

Evaluation Metrics
To survey the execution of the unified learning calculations, we utilized standard
measurements such as exactness, accuracy, review, and F1 score [33]. The assess-
ment was conducted on a separate test dataset, guaranteeing fair experiences in the
model’s generalization capabilities.

Experimental Setup
The tests were conducted on a simulated IoT environment utilizing Python and
TensorFlow. The IoT gadgets were imitated with different computing capabilities,
264 Artificial Intelligence Using Federated Learning

FIGURE 13.8 ClientUpdate model of DPFL

TABLE 13.6
Various values of parametric evaluations in DPFL
Parameter Value
Privacy Budget 1.0
Number of Rounds (T) 25

and the combined learning calculations were actualized utilizing fitting libraries and
systems [34].

Statistical Analysis
Statistical importance tests, such as t-tests, were utilized to approve the execution
contrasts between the unified learning calculations [35]. Furthermore, the privacy-
preserving viewpoints were assessed by analyzing the sum of data spillage and dem-
onstrating utility.

Experiments
Experimental Setup
To assess the execution of the unified learning calculations in privacy-preserving
machine learning for IoT systems, a comprehensive set of tests was conducted. The
tests centered on surveying show accuracy, security conservation, and computa-
tional productivity. The recreated IoT environment included different gadgets with
IOT Networks 265

FIGURE 13.9 A federated graph neural network framework for privacy-preserving

personalization

shifting computational capacities, reflecting real-world scenarios [36]. The com-

bined learning calculations, specific Federated Averaging (FedAvg), Homomorphic
Encryption-based Unified Learning, Secure Aggregation, and Differential Protection
in Combined Learning was executed utilizing Python and TensorFlow as depicted
in Figure 13.9 [37].

Evaluation Metrics
The tests utilized a run of standard assessment measurements to evaluate the
execution of the combined learning calculations. Key measurements included
accuracy, precision, recall, and F1 score, giving an all-encompassing see of the
models’ prescient capabilities as depicted in Figure 13.10. Privacy-related mea-
surements, such as data spillage and differential security ensures, were also mea-
sured [38]. Moreover, computational measurements, counting preparing time,
and communication overhead, were considered to assess the productivity of the
calculations.

Comparison With Related Work

To contextualize our results, a comparative investigation was conducted with exist-
ing studies in privacy-preserving machine learning for IoT systems. Notable related
work incorporates, where a combined learning approach with accentuation on pro-
tection was proposed, which investigated homomorphic encryption in combined
learning [39]. Our tests point to constructing upon and amplifying the discoveries of
266 Artificial Intelligence Using Federated Learning

FIGURE 13.10 Federated learning framework with differential privacy update

these studies, advertising bits of knowledge into the comparative adequacy of differ-
ent unified learning calculations in an IoT setting.

Results
Model Accuracy
Table 13.7 presents the precision measurements of the unified learning calculations
on a test dataset. The models were prepared for a settled number of rounds, and the
precision was assessed on a partitioned test set to degree generalization execution.
The results demonstrate that FedAvg accomplished the most noteworthy precision,
exhibiting its adequacy in collaborative learning. In any case, Secure Aggregation
and Differential
Security illustrated competitive precision levels, emphasizing their utility in
scenarios where protection conservation is fundamental [40]. Figure 13.11 depicts
Privacy-preserving machine learning and multi-party computation.

Privacy Preservation
Security conservation could be a basic angle of unified learning in IoT systems.
Table 13.8 gives an outline of privacy-related measurements, counting data leakage,
and the level of differential protection accomplished by each calculation.
Homomorphic Encryption and Differential Privacy algorithms show moo data
leakage, guaranteeing that the prepared models don’t incidentally uncover points of
interest around personal information focuses [26]. Differential Privacy, in particu-
lar, accomplished a security parameter (ε) of 1.0, showing a tall level of protection
conservation.
IOT Networks 267

TABLE 13.7
Various Algorithms with Accuracies
Algorithm Accuracy (%)
FedAvg 92.5
Homomorphic Encryption 88.2
Secure Aggregation 91.8
Differential Privacy 89.7

FIGURE 13.11 Privacy-preserving machine learning and multi-party computation

TABLE 13.8
Various Algorithms with Information Leakage and Differential Privacy
Algorithm Information Leakage Differential Privacy (ε)
FedAvg Moderate Not applicable
Homomorphic Encryption Low 2.5
Secure Aggregation Minimal Not applicable
Differential Privacy Minimal 1.0
268 Artificial Intelligence Using Federated Learning

Computational Efficiency
Efficient show preparation and communication are significant for combined learning
in resource-constrained IoT situations.
Table 13.9 traces the computational proficiency measurements, counting prepar-
ing time, and communication overhead.
Secure Aggregation illustrated low communication overhead, making it reason-
able for situations with restricted transmission capacity. In any case, Holomorphic
Encryption exhibited high communication overhead due to the encryption and
unscrambling forms. FedAvg and Differential Privacy fell inside direct communica-
tion overhead levels, striking an adjustment between effectiveness and protection.
Figure 13.12 depicts federated learning and its role in the privacy preservation of
IoT devices.

TABLE 13.9
Various Algorithms with Training Time and Communication Overhead
Algorithm Training Time (s) Communication Overhead
FedAvg 120 Moderate
Holomorphic Encryption 280 High
Secure Aggregation 150 Low
Differential Privacy 200 Moderate

FIGURE 13.12 Federated Learning and Its Role in the Privacy Preservation of IoT Devices
IOT Networks 269

Calculation Parameters
Here, in Figure 13.13 calculation parameters with comparisons are shown.
Accuracy, Recall, Precision, Specificity, and Sensitivity are the classification param-
eters. Consider these parameters to analyze the result.

Experimental Result
Three Python libraries Sklearn, NumPy, and Tensor Flow are employed for data
analysis, mathematical operations, categorization, prediction, and the creation of
data flow graphs. The estimated accuracy is 98%. Precision, Recall, and F1- Score is
98%, 90.7%, and 31% respectively.

Performance Percentage
Table 13.10 shows the performance percentage of various algorithms with various
parameters. Figure 13.14 depicts the performance comparison of SVM and Federated

FIGURE 13.13 Comparison Parameters

TABLE 13.10
Performance Percentage
Logistic Federated
Decision Tree Regression Random Forest SVM Learning
Accuracy 97.94 96.94 97.95 98.9 99.3
Precision 97.61 98.59 94.31 92.7 98
Recall 46.06 48.44 67.68 64.1 90.7
F1-Score 62.59 64.96 78.81 64.1 31
270 Artificial Intelligence Using Federated Learning

FIGURE 13.14 SVM, Federated learning

FIGURE 13.15 Random forest, SVM, Federated learning

learning. Figure 13.15 depicts the performance comparison of random forest, SVM,
and Federated learning. Figure 13.16 depicts the performance comparison of logis-
tic regression, random forest, SVM, and Federated learning. Figure 13.17 depicts
the performance comparison of decision tree, random forest, SVM, logistic regres-
sion, and federated learning:
IOT Networks 271

FIGURE 13.16 Logistic regression, Random forest, SVM, Federated learning

FIGURE 13.17 Decision tree, Logistic regression, Random forest, SVM, Federated learning

Discussion
The results highlight the trade-offs among the combined learning calculations in
terms of demonstrated precision, security conservation, and computational profi-
ciency. FedAvg exceeded expectations in exactness but needed unequivocal security
272 Artificial Intelligence Using Federated Learning

conservation instruments [27]. Homomorphic Encryption and Differential Privacy,

on the other hand, illustrate robust
Security ensures but with expanded computational requests. Secure Aggregation
rose as a promising compromise, accomplishing competitive precision while mini-
mizing data spillage and communication overhead. This renders Secure Aggregation
very practical for IoT scenarios that need privacy-preserving and collaborative inter-
actions [13]. Comparing the arises with related work revealed that FedAvg and
Secure Aggregation consistently outrun existing methods in precision and protec-
tion savings. The integration of Holomorphic Encryption and Differential Privacy
algorithms improved the comparison, indicating the feasibility of privacy-preserving
combined learning in various types of IoT models.

Conclusion
To sum up, the study of the work “Federated Learning for Privacy-Preserving
Machine Learning in IoT Systems” has offered valuable insights into the convo-
luted interplay between collaborative learning, safety preservation, and efficiency
in the context that is the Internet of Things (IoT) situations. All the unified learn-
ing algorithms, including Combined Averaging, homomorphic encryption-based
Federated Learning, Secure Aggregation, and Differential Security in Combined
Learning were carefully assessed and compared. The tests revealed that while
FedAvg outperformed expectations in terms of accuracy, algorithms such as Secure
Aggregation trade between accuracy and security preservation, thus ranking as
promising candidates for privacy-sensitive IoT applications. Comparative analysis
with related work showed the relevance of the research in the broader picture of
machine learning, blockchain integration, and IoT. Combined learning integrated
with IoT as studies by various authors showed emphasis on the versatility of these
improvements in various fields such as smart cities, healthcare, and remote com-
munication. The introduced systems and calculations here help in addressing the
issues dependent on heterogeneity, non-IID data dissemination, and asset requests
in IoT settings.
In addition, the investigation contextualized itself inside the progressing insightful
discourse on privacy-preserving unified learning, drawing associations with related
considerations that investigated inventive arrangements and applications. The union
of discoveries from different ponders contributes to an all-encompassing understand-
ing of combined learning’s potential, challenges, and arrangements, progressing the
collective information base within the crossing point of machine learning and IoT.
As the computerized scene proceeds to advance, the investigation underscores the
significance of privacy-preserving unified learning as a significant worldview for
dependable and compelling data-driven decision-making in IoT biological systems.
Online fraud detection is currently a worldwide epidemic. When fraudsters produce
erratic patterns that resemble the original, a more effective method of identifying
online frauds while protecting users’ privacy is required. In this case, online scams
are detected while maintaining privacy using Deep Learning, Machine Learning,
and Federated Learning techniques. Compare the existing techniques with Federated
Learning with MLP, where the accuracy has been increasing.
IOT Networks 273

SVM is overfitting for large datasets, which is used in this research. In future
work, proposed split learning can be implemented and tested with different machine
learning methods. Future endeavors in this space seem constructed upon these
experiences, refining calculations, and systems for particular IoT applications and
addressing developing challenges within the ever-evolving scene of interconnected
gadgets.

REFERENCES
[1] Alazab, A., Khraisat, A., Singh, S., & Jan, T. (2023). Enhancing privacy-preserving
intrusion detection through federated learning. Electronics, 12(16), 3382.
[2] Chai, J., Li, J., Wei, M., & Zhu, C. (2023). Blockchain managed federated learning
for a secure IoT framework. EURASIP Journal on Wireless Communications and
Networking, 2023(1), 100.
[3] Du, W., Li, M., Wu, L., Han, Y., Zhou, T., & Yang, X. (2023). A efficient and robust pri-
vacy-preserving framework for cross-device federated learning. Complex & Intelligent
Systems, 9(5), 4923–4937.
[4] Valencia-Arias, A., González-Ruiz, J. D., Verde Flores, L., Vega-Mori, L., Rodríguez-
Correa, P., & Sánchez Santos, G. (2024). Machine learning and blockchain: A biblio-
metric study on security and privacy. Information, 15(1), 65.
[5] Ghadi, Y. Y., Mazhar, T., Shah, S. F. A., Haq, I., Ahmad, W., Ouahada, K., & Hamam,
H. (2023). Integration of federated learning with IoT for smart cities applications, chal-
lenges, and solutions. PeerJ Computer Science, 9, e1657.
[6] Yu, X., Tang, D., & Zhao, W. (2023). Privacy-preserving cloud-edge collaborative learn-
ing without trusted third-party coordinator. Journal of Cloud Computing, 12(1), 19.
[7] Zeng, Q., Lv, Z., Li, C., Shi, Y., Lin, Z., Liu, C., & Song, G. (2023). Fedprols: Federated
learning for iot perception data prediction. Applied Intelligence, 53(3), 3563–3575.
[8] Zhang, K., Cai, Z., & Seo, D. (2023). Privacy-preserving federated graph neural network
learning on non-iid graph data. Wireless Communications and Mobile Computing, 2023.
[9] Zhao, R., Xie, Y., Cheng, H., Jia, X., & Shirazi, S. H. (2023). ePMLF: Efficient and pri-
vacy-preserving machine learning framework based on fog computing. International
Journal of Intelligent Systems, 2023.
[10] Almalki, J., Alshahrani, S. M., & Khan, N. A. (2024). A comprehensive secure system
enabling healthcare 5.0 using federated learning, intrusion detection and blockchain.
PeerJ Computer Science, 10, e1778.
[11] Al Asqah, M., & Moulahi, T. (2023). Federated learning and blockchain integration for
privacy protection in the Internet of Things: Challenges and solutions. Future Internet,
15(6), 203.
[12] Butt, M., Tariq, N., Ashraf, M., Alsagri, H. S., Moqurrab, S. A., Alhakbani, H. A. A., &
Alduraywish, Y. A. (2023). A fog-based privacy-preserving federated learning system
for smart healthcare applications. Electronics, 12(19), 4074.
[13] Sathiamoorthy, A., Mithusan, S., Rathnayaka, R. M. L. R., Kajenthiran, S., Hansika,
M. M., & Pandithage, D. (2023). StreamSafe: Improving QoS and security in IoT net-
works. International Research Journal of Innovations in Engineering and Technology,
7(11), 170.
[14] Kaushik, A., Gahletia, S., Garg, R. K., Sharma, P., Chhabra, D., & Yadav, M. (2022,
December). Advanced 3D body scanning techniques and its clinical applications.
In 2022 International Conference on Computational Modelling, Simulation and
Optimization (ICCMSO) (pp. 352–358). IEEE.
274 Artificial Intelligence Using Federated Learning

[15] Singh, K., Singh, Y., Barak, D., & Yadav, M. (2023). Comparative performance analy-
sis and evaluation of novel techniques in reliability for Internet of Things with RSM.
International Journal of Intelligent Systems and Applications in Engineering, 11(9s),
330–341.
[16] Singh, K., Singh, Y., Barak, D., & Yadav, M. (2023). Evaluation of designing tech-
niques for reliability of Internet of Things (IoT). International Journal of Engineering
Trends and Technology, 71(8), 102–118.
[17] Singh, K., Singh, Y., Barak, D., Yadav, M., & Özen, E. (2023). Parametric evalua-
tion techniques for reliability of Internet of Things (IoT). International Journal of
Computational Methods and Experimental Measurements, 11(2), 123–134.
[18] Singh, K., Singh, Y., Barak, D., & Yadav, M. (2023). Detection of lung cancers from
CT images using a deep CNN architecture in layers through ML. In AI and IoT-Based
Technologies for Precision Medicine (pp. 97–107). IGI Global.
[19] Singh, K., Yadav, M., Singh, Y., & Barak, D. (2023). Reliability techniques in IoT envi-
ronments for the healthcare industry. In AI and IoT-Based Technologies for Precision
Medicine (pp. 394–412). IGI Global.
[20] Sharma, H., Singh, K., Ahmed, E., Patni, J., Singh, Y., & Ahlawat, P. (2021). IoT based
automatic electric appliances controlling device based on visitor counter. https://doi.
org/10.13140/RG,2(30825.83043).
[21] Bhatia, S., Goel, A. K., Naib, B. B., Singh, K., Yadav, M., & Saini, A. (2023,
July). Diabetes prediction using Machine learning. In 2023 World Conference on
Communication & Computing (WCONF) (pp. 1–6). IEEE.
[22] Chen, Z., Cui, H., Wu, E., & Yu, X. (2023). Computation and communication effi-
cient adaptive federated optimization of federated learning for Internet of Things.
Electronics, 12(16), 3451.
[23] Han, Y., & Zhu, X. (2023). Enhancing throughput using channel access priorities in
frequency hopping network using federated learning. EURASIP Journal on Wireless
Communications and Networking, 2023(1), 101.
[24] Du, W., Wang, Y., Meng, G., & Guo, Y. (2024). Privacy-preserving vertical federated
KNN feature imputation method. Electronics, 13(2), 381.
[25] El-Gendy, S., Elsayed, M. S., Jurcut, A., & Azer, M. A. (2023). Privacy preservation
using machine learning in the Internet of Things. Mathematics, 11(16), 3477.
[26] Mouhni, N., Elkalay, A., Chakraoui, M., Abdali, A., Ammoumou, A., & Amalou, I.
(2022). Federated learning for medical imaging: An updated state of the art. Ingenierie
des Systemes D’Information, 27, 143–150.
[27] Pinto Neto, E. C., Sadeghi, S., Zhang, X., & Dadkhah, S. (2023). Federated rein-
forcement learning in IoT: Applications, opportunities and open challenges. Applied
Sciences, 13(11), 6497.
[28] www.kaggle.com/datasets/rupakroy/online-payments-fraud-detection-dataset
[29] Javed, A., Awais, M., Shoaib, M., Khurshid, K. S., & Othman, M. (2023). Machine
learning and deep learning approaches in IoT. PeerJ Computer Science, 9, e1204.
[30] Karras, A., Giannaros, A., Theodorakopoulos, L., Krimpas, G. A., Kalogeratos, G.,
Karras, C., & Sioutas, S. (2023). FLIBD: A federated learning-based IoT big data man-
agement approach for privacy-preserving over apache spark with FATE. Electronics,
12(22), 4633.
[31] Kea, K., Han, Y., & Kim, T. K. (2023). Enhancing anomaly detection in distributed power
systems using autoencoder-based federated learning. PLoS ONE, 18(8), e0290337.
[32] Munawar, A., & Piantanakulchai, M. (2024). A collaborative privacy-preserving
approach for passenger demand forecasting of autonomous taxis empowered by feder-
ated learning in smart cities. Scientific Reports, 14(1), 2046.
IOT Networks 275

[33] Muthukumar, V., Sivakami, R., Venkatesan, V. K., Balajee, J., Mahesh, T. R., Mohan,
E., & Swapna, B. (2023). Optimizing heterogeneity in IoT Infra using federated learning
and blockchain-based security strategies. International Journal of Computers
Communications & Control, 18(6).
[34] Peyvandi, A., Majidi, B., Peyvandi, S., & Patra, J. C. (2022). Privacy-preserving feder-
ated learning for scalable and high data quality computational-intelligence-as-a-service
in Society 5.0. Multimedia Tools and Applications, 81(18), 25029–25050.
[35] Qin, J., Zhang, X., Liu, B., & Qian, J. (2023). A split-federated learning and edge-
cloud based efficient and privacy-preserving large-scale item recommendation model.
Journal of Cloud Computing, 12(1), 57.
[36] Rashid, M. M., Khan, S. U., Eusufzai, F., Redwan, M. A., Sabuj, S. R., & Elsharief,
M. (2023). A federated learning-based approach for improving intrusion detection in
industrial internet of things networks. Network, 3(1), 158–179.
[37] Song, Z. (2020, June). A data mining based fraud detection hybrid algorithm in E-bank.
In 2020 International Conference on Big Data, Artificial Intelligence and Internet of
Things Engineering (ICBAIE) (pp. 44–47). IEEE.
[38] Rodríguez, E., Otero, B., & Canal, R. (2023). A survey of machine and deep learning
methods for privacy protection in the internet of things. Sensors, 23(3), 1252.
[39] Hsu, H. Y., Keoy, K. H., Chen, J. R., Chao, H. C., & Lai, C. F. (2023). Personalized fed-
erated learning algorithm with adaptive clustering for non-IID IoT data incorporating
multi-task learning and neural network model characteristics. Sensors, 23(22), 9016.
[40] Kolhar, M., & Aldossary, S. M. (2023). Privacy-preserving convolutional Bi-LSTM
network for robust analysis of encrypted time-series medical images. AI, 4(3), 706–720.
 14 Federated Query
Processing for Data
Integration Using
Semantic Web
Technologies
A Review
Nidhi Gupta, Pawan Verma, Monali Gulhane,
Nitin Rakesh, and Ahmed A. Elngar

INTRODUCTION
Data science and analytics are a growing interest of many organizations. They require
collecting, processing, and analyzing the data to generate meaningful insights. The
first and most significant step is to collect data from various sources. Data integra-
tion refers to combining data from multiple data sources to get a unified view of
data. It extracts the data, transforms it, and loads it in a central place using extract,
transform, and load (ETL) tools. To effectively handle heterogeneous databases, we
need tools that can address the specific challenges and requirements of each system,
rather than relying solely on a centralized ETL approach.
The semantic web helps to share data and promote interoperability between vari-
ous systems. The term semantic web refers to linked data on the web. The aim of
linked open data (LOD) is to globally access the data on the web. With the success
of LOD for open data, semantic technologies are also being used to access organi-
zational data. The semantic web uses the resource description framework (RDF)
to represent the data. It facilitates data interchange on the web. In comparison to
relational databases, linked data represents the resources using a uniform resource
indicator (URI) that helps to represent a resource uniquely.
Relational databases are widely used for storing data. SQL has grown as a pow-
erful query language for querying relational databases and integrating data from
multiple sources. Relational databases impose strict schema on databases which are
not suited for applications that require frequent schema changes. The data generated
from multiple sources such as wearable devices and other IoT applications does not
adhere to strict schemas, thus moving towards flexible databases like MongoDB for

276 DOI: 10.1201/9781003482000-14

Federated Query Processing for Data Integration 277

storing formats such as JSON and XML. Therefore, graph databases like Virtuoso
and Graph-DB are gaining popularity to integrate multiple data sources into an inte-
grated graph. They utilize the RDF framework to represent the data and perform
queries for data access.
Query processing provides data access to large volumes of data stored in multiple
systems. There are two approaches for query processing over the web of linked and
distributed data sources: link traversal and query federation. Link traversal (Hartig
et al., 2009) such as SQUIN (Squin, 2013) searches relevant data by following links
between RDF data sources. It does not require the prior knowledge of data sources.
The query executes by binding the intermediate results with common variables.
The link traversal discovers the data source at runtime, hence providing up-to-date
results. The major weakness of this approach is that the wrong selection of a starting
point for traversing the link would lead to a large number of intermediate results and
could not guarantee complete results. Another approach is query federation, which
inputs a query and distributes it to relevant data sources. Federated query processing
for data integration is a crucial concept in the field of data management and informa-
tion retrieval. It refers to the process of querying and retrieving data from multiple
distributed or heterogeneous data sources and integrating the results into a unified
response. This approach is employed when organizations or systems need to access
and combine data from various repositories, databases, or systems to derive mean-
ingful insights or to meet specific information needs.
A federated engine acts as a mediator to distribute subqueries and collect the results.
It utilizes data descriptions to find the relevant data sources and provides efficient query
plans to minimize query execution time and maximize result completeness. Federated
query processing is advantageous for result completeness and providing up-to-date
results. Therefore, this chapter focuses on federated query processing and reviews the
work carried out on optimizing federated queries for efficient data access.
The chapter reviews the work that has been performed to integrate heterogeneous
data sources using semantic technologies and its access via federated query process-
ing. It presents the various approaches used for federated query processing and com-
pares various federated engines based on these approaches. Further, it discusses the
target problems and their proposed solutions in health data integration and federated
query processing.
In the second section, we provide a concise overview of the background and rel-
evant research. The third section presents ontology-based data integration for dis-
tributed data access. The fourth section explores the federated approach to query
distributed data sources. The fifth section provides different challenges and reviews
the target problems on query execution and their solutions. Finally, the sixth section
concludes with future work.

BACKGROUND AND RELATED SURVEYS

Data Integration and Interoperability
Data integration is the process of combining data from different data sources to provide
a unified view of data that facilitates data access through a single query interface. The
278 Artificial Intelligence Using Federated Learning

integration of data from heterogeneous data sources is a challenge for organizations, as

data at distinct sources differs in schemas, data formats, and models used. Data integra-
tion requires interoperability for data access. Interoperability refers to the ability of sys-
tems to share data. It is significant to achieve this, specifically in a heterogeneous data
environment. A study by Gupta and Gupta (2019) discussed syntactical and semantic
interoperability. Syntactic interoperability requires standardizing health data structures.
Various health standards have been developed for data storage and exchange such as
HL7 FHIR and OpenEHR. Lack of adoption of standards by health providers results in
heterogeneous systems and leads to data interoperability problems.
Semantic interoperability ensures the use of various tools and mappings to ensure
meaningful interpretations. It uses appropriate data representation and mappings
that allow multiple distinct organizations to share and integrate data for both stan-
dard and non-standard electronic health record (EHR) data formats. The work car-
ried out studied the state-of-art approaches of federated query engines that perform
integration of data sources in non-standard data formats.

Types of Data Integration Systems

The integration of data is categorized in two different ways in accordance with data
storage and query processing. These are:

1. Centralized: Centralized data integration involves gathering and unifying

data from multiple disparate sources and storing it in a centralized location.
2. Concise: Centralized data integration consolidates data into a single reposi-
tory. The data is accessed from distinct data sources and put into a single
RDF data store. In this, data processing and querying takes place at one
central place. The centralized approach is beneficial in terms of efficient
query processing due to centralized and optimized index structures. It
requires making copies of data for central storage, thus making data man-
agement and update difficult.
3. Distributed system: A distributed system consists of data at various physi-
cal locations. Distributed systems may have federated or peer-to-peer data
success. The details of each are:
• Federated: A federated database system is a special distributed database
system that provides an alternative to a centralized data system. Feder-
ated data integration refers to the consolidation of distributed, autono-
mous data to a virtual unified data model using a central mediator. The
mediator maintains an index that stores the statistical information of data
sources, which enables relevant selection of data sources for an input
query. Data is accessed from distinct data sources and does not require
organizations to move their data, thus maintaining data privacy.
• Peer to peer: The peer-to-peer system consists of various peers or
nodes, each cooperating and having its own data. The data and index are
maintained in a distributed fashion. Thus, there is no central mediator,
and the query is processed by each node. These systems are flexible, as
they can handle changes in peer configurations.
Federated Query Processing for Data Integration 279

4. Semantic web technologies: Semantic technologies are used to add mean-

ing and context to data, making it possible to understand relationships and
concepts even when data comes from diverse sources. Some of the widely
used technologies used for data integration are:
a. URI: A uniform resource identifier, as the name implies, uniquely iden-
tifies any resource on the web. A Uniform Resource Locator (URL) is a
subset of a URI. In addition to resource identification, URLs gives the
way to access the resource; for example, http://ex.org identifies a URI
and is accessed using HTTP protocol (Masinter et al., 2005).
b. Resource Description Framework: It is a World Wide Web Consortium
(W3C) standard framework to represent information about the resources
in a machine-readable format. RDF data is expressed in the form of
triple patterns. A triple consists of <subject, predicate, object>. The sub-
ject represents a resource as URI, the predicate describes the relation
between resources that are represented as a property of the resource, and
the object signifies the value of the property. For example: triplet <pid20
temp “99”> represents the temperature observation value of patient id
‘pid20’ as ‘99’.
c. RDF Schema (RDFS): An RDFS defines the vocabulary of the RDF
dataset. It provides mechanisms to group the related resources called
classes, and members of the classes are instances of that class. The
rdf: type property states that the resource is an instance of the class.
RDFS is written in RDF using various constructs to define RDF such
as rdf: type, rdf: class, rdf:property, rdf:domain, and rdf:range. The
W3C provides the details of RDF schema properties (Faheem et al.,
2018).
d. OWL (Web Ontology Language): Ontology language is also a data
modeling language formally used to describe a resource. OWL pro-
vides much more vocabulary of data models and is particularly used
in the automatic reasoning process. Besides RDFS properties, OWL
also defines equivalence across databases using owl:same. By con-
structing annotations, we can establish connections and define rela-
tionships between different standard ontologies.e. RDF formats:
An RDF document is written using various file formats called RDF
serialization format. The various formats are Turtle (with extension
is.ttl), Notation 3 (with extension.n3), and N-triples (with extension.
nt). JSON-LD (with extension.jsonld) is the JSON syntax of an RDF
document, and RDF/XML is an XML syntax of an RDF document
(Beckett & McBride, 2004)
f. SPARQL: The query language for RDF data stores is SPARQL. It con-
sists of a basic graph pattern (BGP) which has conjunctive triple pat-
terns. The triples use variable names (prefixed by?) to retrieve the values
from the dataset. The triples in BGP are matched with the RDF triples
and return variable bindings. Listing 14.1 shows the SPARQL query for
retrieving the value of blood pressure (systolic, diastolic) and tempera-
ture of all patients with given filter value ranges.
280 Artificial Intelligence Using Federated Learning

LISTING 14.1
SPARQL Query

Select ?pid ?sys ?dys ?temp

Where{
?s pid ?pid .
?s sys ?sys. FILTER (?sys >100 && ?sys<120).
?s dys ?dys FILTER (?dys < 70).
?s temp ?temp
}

Related Surveys
This chapter focuses on two main aspects, semantic data integration and federated
query processing.
In the literature review, Peng et al. (2020) investigated existing problems in the
integration of health data and different approaches to its integration. It highlights
the open problems in integrating aggregated health data and advocates for the use of
semantic technologies with web application interfaces to overcome the challenges.
The research work by Wylot et al. (2018) presented biological database modeling
with semantic technologies. It focuses on ontology-based integration of biological
data and outlines the semantic technologies used for its integration.
The research work by Asfand-e-Yar and Ali (2020) focused on ontology-based
semantic integration of heterogenous databases of the same domain using data trans-
lation and query translation techniques.
The survey by Ouzzani and Bouguettaya (2004) focused on fundamental problems
in efficient query processing. It also discusses various optimization techniques over
web data integration systems and frameworks to evaluate them. The authors in Ali
et al. (2017) experimentally evaluate the various query federated engines on the effi-
ciency of source selection, data partition, answer completeness, and query runtime.
Oguz et al. (2015) presented a qualitative survey for federated query processing on
linked data. It provides details of different approaches to federated query processing
steps such as source selection and query optimization and compares various federated
engines on those approaches. The authors in Sima et al. (2019) provide a comprehen-
sive study of RDF data storage and management systems. It provides different RDF
storage mechanisms, indexing, architecture, and query execution methods.
Several studies conducted either data integration or federated queries, but very
limited surveys cover both aspects of data integration and federated query process-
ing. Thus, the work carried out surveys the state-of-art approaches in the area of data
integration and federated query processing. It identified problems faced during inte-
gration and querying heterogeneous data sources and also discussed their solutions.

ONTOLOGY-BASED DATA INTEGRATION

FOR DISTRIBUTED DATA ACCESS
Ontology-based data integration (OBDA) system consists of triple <O, S, M>, where
O is a domain ontology, S is the source schema, and M defines the mappings between
Federated Query Processing for Data Integration 281

the source and target domain ontology. OBDA uses a global view to query distinct
heterogeneous data sources using an ontology, thereby masking the actual imple-
mentation details of data sources. One of the major challenges for data integration
is data heterogeneity. The problem of heterogeneity is resolved using semantic tech-
nologies in one of the following ways:

1. Materialized approach: Data in the same format can be achieved by con-

version of non-RDF data to RDF data format using tools such as Open
Refine (Donnelly, 2014). These tools clean the data, transform it into the
desired data format, and link it with other web services. Since data is main-
tained in a single place, this approach is efficient in terms of query process-
ing. However, its maintenance is a costly process, as it requires copies of
data to be stored at central storage and periodic updating of data.
2. Virtual approach: The cost of materialization of databases is removed
by using the virtual view of databases. It creates an RDF view over non-
RDF data sources. The mappings are defined to create an RDF view; for
instance, relational data is viewed as RDF using R2RML mappings, while
non-relational data gives RDF views using RML mappings. Heterogenous
data sources are queried over integrated RDF views and queried at their own
place.

In recent decades, numerous OBDA mapping languages have emerged. To estab-

lish a common framework for transforming relational data into RDF, the RDB2RDF
W3C working group was created. Direct mapping (Sequeda et al., 2012) and
R2RML (Priyatna et al., 2015) are two prominent examples of these mapping lan-
guages. Direct mapping transforms RDB to RDF by a predefined procedure without
user intervention, while R2RML is a customizable language to map RDB to RDF.
R2RML mappings offer a more flexible approach than direct or indirect mappings.
They allow us to map database schemas to standard RDF vocabularies, providing a
consistent and global view for SPARQL queries. This flexibility is especially useful
when dealing with database schemas that cannot be easily modified. R2RML output
may be materialized RDF or RDF virtual view. Later, other mapping languages were
proposed for mapping non-relational databases, such as RML (Dimou et al., 2014),
proposed to map non-relational databases (JSON, CSV, and XML data sources) to
RDF and xR2RML (Michel et al., 2016) for MongoDB databases. The mappings
achieve semantic interoperability that helps to perform integrated data access by
exposing heterogeneous data as SPARQL endpoints.
Integration of data requires aggregation from various data sources to form a
linked graph. Figure 14.1 shows the architecture for distributed data access from
RDF and non-RDF data sources. It consists of four layers:

1. Data layer: This layer consists of distinct heterogeneous data sources. It

stores the data in different databases: relational, JSON, and RDF. The data
is fetched using an application interface or wrappers. The data retrieved is
aggregated at the integration layer.
2. Data integration layer: The integration layer is responsible for virtual data
integration and federated query processing. The integration layer aggregates
282 Artificial Intelligence Using Federated Learning

the data from multiple data sources. It creates mappings such as R2RML
and RML to translate data sources of distinct formats such as RDB or JSON
to an RDF view. Semantic annotation of data is required with commonly
used vocabularies such as FOAF to make data meaningful and easier to
understand. Federated query processing fetches the data from multiple data
sources by decomposing the input query, selecting the relevant data source
for each subquery preparing an optimized plan and executing it to extract
the required results. The user query is translated to local database queries
by using declarative mappings (RML, R2RML, etc.). The result fetched is
integrated as a result set and sent to the user-interface layer.
3. Semantic layer: It describes the domain ontology to represent various con-
cepts and their relationships. The schema of data sources is aligned to
domain ontology to form a global query.
4. User interface layer: The user interface allows users to input queries to the
federated system and receive the responses.

FIGURE 14.1 Architecture for data integration and querying.

Federated Query Processing for Data Integration 283

FEDERATED QUERY PROCESSING FOR DISTRIBUTED RDF DATA

Federated query engines are widely used in data retrieval and its integration. The
query engines use different approaches to query distributed databases. The basic
steps are query parsing, data source selection, query optimization, and query
execution.
There are two important optimization steps of federated query engines.

Source Selection
Data integration queries retrieve data from multiple data sources; thus, many data
sources are involved in answering an input query, and each source may or may not
contribute to the final answer. Hence, a federated engine is required to search the rel-
evant data sources to answer a given query. It divides the query into triples and finds
the data sources of each query triple. Some of the federated engines use metadata
catalogs. This catalog stores: (1) service descriptions: statistical information of data
sources; (2) a vocabulary of interlinked datasets (VoID) descriptions; that is, it stores
statistical information of data sources along with the metadata of RDF data sources
whose subjects are linked with objects of other RDF data sources; and (3) a list of
predicates at each RDF data source.
Some of the federated engines such as FEDX use indexes to manage data
sources and store them in a cache to avoid repeating the process for the same
input query.

Different Approaches to Data Source Selection

• ASK: The SPARQL ASK query broadcasts the input query to all the
data sources and returns ‘true’ if the data source has the solution to
query; else it returns ‘false’. ASK is generally used for source selec-
tion when the federated engine does not maintain any data for storing
predicate information or the query has a predicate variable. Data source
selection using ASK does not require to maintain data source predicate
information but increases the number of remote requests. Some of the
federated engines that use ASK queries for source selection are FedX
and SPLENDID.
• Predicate-based selection: The federated engine maintains the metadata
catalog, which contains statistical information and predicate information of
each data source. It selects data sources by matching them with the predi-
cates in the metadata catalog. Most of the federated engines use a predicate-
based approach such as ANAPSID, ADERIS, or POLYSTORE.
• Type-based selection: It generally uses type definitions (i.e., rdf:type) of
data sources and matches them with the metadata catalog. SPLENDID also
uses rdf:type information for source selection.
• Rule-based selection: In this method, rules are formed by analyzing the
relationship between query triples. Data source selection is performed by
applying these rules only. WoDQA (Akar et al., 2012) is an example feder-
ated engine that uses rule-based selection.
284 Artificial Intelligence Using Federated Learning

Query Optimization and Execution

A SPARQL query is composed of various triple patterns. Multiple data sources are
relevant to the query. Each triple pattern of the query is executed at the relevant data
sources, and the results are joined to obtain a final result set. However, this method
would lead to a large number of remote HTTP requests and consequently a large
number of local joins. Thus, an efficient query plan to group triple patterns and
execute query groups is the key requirement for successful query execution.

Query Groups
• Exclusive Group: The exclusive group is formed with the triples having the
same and only single data source. The grouping may comprise triples with no
shared variable and may result in redundant intermediate results. Thus, the
variant of exclusive groups is to create different groups for the triple with no
shared variable. owl:sameAs can be used to group triples that have the same
object and predicate, but different subjects, into a single equivalence class
• Predicate-Based Join Group (PBJ): PBJ groups are formed by group-
ing triples that are all evaluated against the same set of data sources. Such
groups are executed on relevant endpoints, and the results obtained are
merged using the SPARQL UNION construct. The POLYSTORE federated
engine uses PBJ groups. PJG is a lightweight index-based approach that
uses predicate information at respective endpoints to form query groups.
Listing 14.2 shows the PBJ groups and explicit SERVICE requests of a
SPARQL query from two different RDF service points: Virtuoso (http://
localhost:8890/sparql) and Apache Jena (http://localhost:3030/vital/query).

LISTING 14.2
SPARQL Query With JOINING PBJ Groups and Explicit SERVICE Endpoints

prefix id: <http://127.0.0.1:3333/>

Select ?pid ?sys ?dys ?temp where
{
{SERVICE <http://localhost:8890/sparql> {?s id:pid ?pid. ?s id:sys ?sys
filter(100<?sys && ?sys<120)}}
UNION
{ SERVICE <http://localhost:3030/vital/query> {?s id:pid ?pid. ?s id:sys ?sys
filter(100<?sys && ?sys<120). ?s id:dys ?dys}
}
{
{SERVICE <http://localhost:8890/sparql> {?s id:pid ?pid. ?s id:temp ?temp}}
}}
Federated Query Processing for Data Integration 285

JOIN selection

The JOIN operation selection is significant for the query optimizer.

A federated engine query plan may have four different kinds of joint
operations:
Bind JOIN: In bind join, the intermediate results of the outer object are used
to filter out the result set from the next join. It executes in the same way as
a nested loop join and is usually efficient when the intermediate results are
smaller in size.
Nested Loop JOIN: In this, the inner relation is compared with each row
of outer relations, and all the bindings that satisfy the join condition are
included in the result set.
Hash Join: It uses a hash table to match two relations. Some federated engines,
such as SPLENDID, use hash join by executing the relations in parallel and
joining them locally using a single hash table.
JOIN Ordering: The goal of JOIN ordering is to optimize the query for result
completeness with minimum execution time. It can be achieved by gener-
ating a minimum-cost query plan. In general, federated engines use three
basic approaches for the computation of query cost:
Statistic-based: This approach uses the data statistics information
that is stored in federated catalogs: predicate information and VoiD
descriptions of data sources. It uses statistics information to com-
pute the cost of the query plan by estimating the number of inter-
mediate results produced with each JOIN operation. The statistical
approaches are often accurate, but data source statistics are not avail-
able each time.
Heuristic-based: Heuristics are the rules that are applied to estimate the
order of JOIN operations in the query. Heuristic-based optimization
estimates the query cost by formulating rules based on the query struc-
ture and properties of the operator used such as type of predicate and
FILTER construct. It does not require overhead to maintain statistical
information of data sources but at the same time provides a near-optimal
query plan.
Hybrid: The hybrid approach makes use of both data source statistics and
heuristics to estimate query cost. It uses minimum statistical informa-
tion along with certain heuristics for cost estimation.

A comparison of state-of-the-art SPARQL federated engines is listed in

Table 14.1. The comparison is performed based on data source selection
approaches, data model supported, handling of unbound predicate queries, the
type of JOIN operation, join-order approach, approach of query group creation,
and cache memory usage. Table 14.2 shows the metrics used to evaluate different
federated engines.
286 Artificial Intelligence Using Federated Learning

TABLE 14.1
Comparison of SPARQL Federated Engines
Federated Source Data Unbound JOIN Query
Engines Selection Model Predicate JOIN Type Ordering Groups Cache
DARQ Service RDF Nested loop, Statistics- Exclusive
(2008) descriptions bind based groups
FEDX SPARQL ASK, RDF  Nested loop, Heuristic- Exclusive 
(2011) local cache bind based groups
SPLENDID VoiD, ASK RDF  Hash, bind Statistics- Exclusive
(2011) based groups
HIBISCUS Data summaries RDF Nested loop, Statistics- Exclusive 
(2014) bind based groups
POLYWEB Data summaries, CSV, RDB,  Nested loop, Heuristic- Predicate-
(2019) ASK RDF bind based based join
group

TABLE 14.2
Evaluation Metrics of Federated Engines
Federated Engines M1 M2 M3 M4 M5 M6
DARQ  
FEDX  
SPLENDID   
HIBISCUS    
POLYWEB     
M1: Number of data sources selected for each triple pattern
M2: Number of ASK requests executed
M3: Average source selection time
M4: Number of results returned by each query
M5: Query planning and optimization time
M6: Response time

CHALLENGES FOR SPARQL QUERY PROCESSING

Some of the challenges faced by query processing for data integration are as follows.

Unbound Predicate Queries

The query with an unbound predicate is less selective in comparison to the subject
and object. Thus, selecting a data source using a predicate as a variable is a difficult
task. Various federated engines such as FEDX and POLYSTORE use ASK requests
to deal with unbound predicate problems.
Federated Query Processing for Data Integration 287

Query Evaluation With OPTIONAL in SPARQL Query

OPTIONAL in a SPARQL query is used to handle missing information. The SPARQL
query execution on RDB data stores in OBDA settings requires the SPARQL query
to translate into SQL queries. OPTIONAL is represented as a LEFT OUTER JOIN
operator in relational databases. It makes the query complex, which is difficult to
optimize. Various solutions are proposed to optimize the size and improve the gener-
ated SQL query structure.

Join Order Optimization

An input query searches and fetches the desired data from multiple SPARQL end-
points. During this process, the intermediate results from one end-point are joined
with the results of the next end-point, and so on. It will lead to a large number of
intermediate results and consequently large search space for performing the JOIN.
Therefore, the join order of the query needs to be optimized to minimize the number
of intermediate results.

Two Main Approaches Are Used to Find the Optimal Join Order
1. Static: In this, the join order plan is optimized before the actual execution
of the query. It is optimized based on the statistical count of data sources
and heuristics.
2. Dynamic: The query join order is optimized during the execution of the
query. It implies that the dynamic approach utilizes the intermediate results
obtained during the search. A dynamic approach may reduce the search
space but may require reconstructing the query plan iteratively using inter-
mediate results.

Cost Estimation of Federated Queries

Cost-based query optimizers estimate intermediate cardinalities to find the best plan.
However, the estimation of link cardinalities is a difficult task. It requires detailed
metadata information of all the data sources, which is sometimes not available.
Heuristics are also proposed to estimate the cost of the queries, but they generally
result in sub-optimal estimations.

Presence of Blank Nodes

Blank nodes represent an anonymous resource that is not assigned any URI. A blank
node is only used as a subject or an object in RDF triple. They are local to an RDF
document; hence a blank node n1 in graph G1 is not the same as blank node n1 in
graph G2. Thus, a challenge occurs when data is fetched from distributed sources and
the intermediate solution binds the blank node while processing the JOIN operation.
The previously mentioned challenges are addressed by various researchers in the
existing literature. Table 14.3 lists some recent target problems with SPARQL feder-
ated query processing and their proposed solutions by various researchers.
288 Artificial Intelligence Using Federated Learning

TABLE 14.3
Target Problems and Their Proposed Solutions in SPARQL Query Processing
on Distributed Data Stores
Year Author Problem Proposed Solution
2023 (Aebeloe et al., Degradation of query performance Achieving faster SPARQL query
2023) due to inaccessibility of data from execution through peer-to-peer
SPARQL endpoints due to high technology
traffic
2022 High communication costs during Algorithm for selecting efficient query
(Lim et al., SPARQL query processing execution path
2022)
2021 (Yang et al., High network costs to transfer data The proposed solution is based on two
2023) over federated databases join algorithms that leverage the
network topology information to
reduce the costs of SPARQL query
execution in a distributed environment
2020 (Naacke & Large index size for storing RDF Optimizes the indexes for RDF data
Curé, 2020) subject, predicate, and object storage by proposing dedicated data
information structures and a data partitioning
approach
2018 (Saleem et al., Estimation of join selectivity for Optimization of SPARQL query
2018) optimized cost based query plan execution by efficient source selection
and join cardinality estimation
2018 (Yannakis High query execution time due to Heuristics-based query reordering
et al., 2018) large number of remote requests to method
data sources
2018 (Xiao et al., Effective query evaluation in Optimization to improve SQL query
2018) presence of OPTIONAL in structure
ontology-based data access.
2017 (Montoya High number of intermediate results Statistical estimation of accurate cost
et al., 2017) while joining data from multiple estimation of federated queries
sources

2015 (Wu et al., Large search space during federated Finding optimal join order to minimize
2015) query execution size of intermediate results
2015 (Atre et al., Restriction of reordering left outer Represent queries in graph of super
2015) joins for optimization of SPARQL nodes and optimize query by applying
OPTIONAL pattern queries the properties of nullification, best
match, and minimality

CONCLUSION
The chapter explored the potential of semantic technologies for data integration and
query processing. This study of data integration and query processing represents
a significant step forward in the quest for a more interconnected, meaningful, and
Federated Query Processing for Data Integration 289

efficient data landscape. It has unveiled various approaches to federated query pro-
cessing that leverage semantic web standards. This work highlights the broader con-
text of the research and recognizes the ongoing efforts in the field to overcome the
challenges associated with data integration and querying. Some of the challenges are
still open for researchers to work on in the future.
In the future, researchers, practitioners, and organizations must continue to
explore and harness the transformative power of semantic web technologies. By
doing so, we can unlock new horizons in knowledge discovery, decision-making,
and data utilization in our increasingly data-driven world.

REFERENCES
Aebeloe, C., Montoya, G., & Hose, K. (2023, January 1). Optimizing SPARQL queries
over decentralized knowledge graphs. Semantic Web, 14(6), 1121–1165. https://doi.
org/10.3233/SW-233438
Akar, Ziya, Halaç, Tayfun, Ekinci, Erdem, & Dikenelli, Oğuz. (2012). Querying the Web
of Interlinked Datasets Using VOID Descriptions. https://api.semanticscholar.org/
CorpusID:8868212
Ali, Hasnain, Ermilov, Ivan, & Ngonga Ngomo, A.-C. (2017). Saleem, Muhammad and Khan,
Yasar and Hasnain. An Evaluation of SPARQL Federation Engines Over Multiple
Endpoints. https://doi.org/10.13140/RG.2.2.25972.45444
Asfand-e-Yar, M., & Ali, R. (2020). Semantic integration of heterogeneous databases of
same domain using ontology. IEEE Access, 8, 77903–77919. https://doi.org/10.1109/
ACCESS.2020.2988685
Atre, M. (2015, May). Left bit right: For SPARQL join queries with OPTIONAL pat-
terns (left-outer-joins). In Proceedings of the 2015 ACM SIGMOD International
Conference on Management of Data (pp. 1793–1808). https://doi.org/10.1145/2723372.
2746483
Beckett, D., & McBride, B. (2004). RDF/XML syntax specification (revised). W3C
Recommendation, 10(2/3).
Dimou, A., Sande, M. V., Colpaert, P., Verborgh, R., Mannens, E., & Walle, R. V. (2014).
RML: A generic language for integrated RDF mappings of heterogeneous data. In
Proceedings of the Workshop on Linked Data on the Web, 1184(January).
Donnelly, F. (2014). Processing Government Data: ZIP Codes, Python, and OpenRefine.
Faheem, M., Sattar, H., Bajwa, I. S., & Akbar, W. (2019). Relational database to resource
description framework and its schema. In Intelligent Technologies and Applications:
First International Conference, INTAP 2018, Bahawalpur, Pakistan, October 23–25,
2018, Revised Selected Papers 1 (pp. 604–617). Springer.
Gupta, N., & Gupta, B. (2019). Big data interoperability in e-health systems. In Data Science
and Engineering (Confluence), (Noida, India) 9th International Conference on Cloud
Computing, 2019 (pp. 217–222). https://doi.org/10.1109/CONFLUENCE.2019.8776621
Hartig, O., Bizer, C., & Freytag, J. C. (2009). Executing SPARQL queries over the web of
linked data. In A. Bernstein (Ed.), ISWC 2009. Lecture Notes in Computer Science,
5823, 293–309. Springer. https://doi.org/10.1007/978-3-642-04930-9_19
Lim, J., Kim, B., Lee, H., Choi, D., Bok, K., & Yoo, J. (2022). An efficient distributed
SPARQL query processing scheme considering communication costs in spark environ-
ments. Applied Sciences, 12(1), 122. https://doi.org/10.3390/app12010122
Masinter, L., Berners-Lee, T., & Fielding, R. T. (2005). Uniform resource identifier (URI):
Generic syntax. Network Working Group. Fremont, CA.
290 Artificial Intelligence Using Federated Learning

Michel, F., Faron-Zucker, C. F., & Montagnat, J. (2016, April). A generic mapping-based query
translation from SPARQL to various target database query languages. In Proceedings
of the 12th International Conference on Web Information Systems and Technologies
(WEBIST2016), Italy, 2 (pp. 147–158). https://doi.org/10.5220/0005905401470158
Montoya, G., Skaf-Molli, H., & Hose, K. (2017). The Odyssey approach for optimizing
federated SPARQL queries. In The Semantic Web–ISWC. Proceedings, Part I: 16th
International Semantic Web Conference, Vienna, Austria, October 21–25, 2017, 16
(pp. 471–489). Springer International Publishing.
Naacke, H., & Curé, O. (2020). On distributed SPARQL query processing using triangles of
RDF triples. Open Journal of Semantic Web, 7, 17–32.
Oguz, D., Ergenc, B., Yin, S., Dikenelli, O., & Hameurlain, A. (2015). Federated
query processing on linked data: A qualitative survey and open challenges.
Knowledge Engineering Review, 30(5), 545–563. https://doi.org/10.1017/
S0269888915000107
Ouzzani, M., & Bouguettaya, A. (2004). Query processing and optimization on the web.
Distributed and Parallel Databases, 15(3), 187–218. https://doi.org/10.1023/B:DAPD.000001
8574.71588.06
Peng, C., Goswami, P., & Bai, G. (2020, September). A literature review of current tech-
nologies on health data integration for patient-centered health management. Health
Informatics Journal, 26(3), 1926–1951. https://doi.org/10.1177/1460458219892387.
Epub December 30, 2019. PubMed: 31884843
Priyatna, F., Alonso-Calvo, R., Parasio-Medina, S., Padron-Sanchez, G., & Corcho, O. (2015,
December). R2RML-based access and querying to relational clinical data with morph-RDB.
In 8th International Conference on Semantic Web Applications and Tools for Life Sciences,
Cambridge United Kingdom (pp. 142–151). https://ceur-ws.org/Vol-1546/paper_21.pdf
Saleem, M., Potocki, A., Soru, T., Hartig, O., & Ngomo, A. N. (2018). CostFed: Cost-based
query optimization for SPARQL endpoint federation. Procedia Computer Science, 137,
163–174. https://doi.org/10.1016/j.procs.2018.09.016
Sequeda, J. F., Arenas, M., & Miranker, D. P. (April 2012). On directly mapping relational
databases to RDF and OWL. In Proceedings of the 21st International Conference on
World Wide Web (WWW 2012) (pp. 649–658). Association for Computing Machinery.
https://doi.org/10.1145/2187836.2187924
Sima, A. C., Stockinger, K., de Farias, T. M., & Gil, M. (2019). Semantic integration and
enrichment of heterogeneous biological databases. In M. Anisimova (Ed.), Evolutionary
Genomics. Methods in Molecular Biology, 1910, 655–690. Humana Press. https://doi.
org/10.1007/978-1-4939-9074-0_22
Squin, H. O. (2013). A traversal based query execution system for the web of linked
data. In Proceedings of the 2013 ACM SIGMOD International Conference on
Management of Data 2013 Jun 22 (pp. 1081–1084). https://api.semanticscholar.org/
CorpusID:10158072
Wu, H., Yamaguchi, A., & Kim, J. D. (2015). Dynamic join order optimization for SPARQL
endpoint federation. In SSWS@ISWC (pp. 48–62). https://api.semanticscholar.org/
CorpusID:16681995
Wylot, M., Hauswirth, M., Cudre-Mauroux, P., & Sark, S. (2018). RDF data storage and
query processing schemes: A survey. ACM Computing Surveys, 51, 1–36. https://doi.
org/10.1145/3177850
Xiao, G., Kontchakov, R., Cogrel, B., Calvanese, D., & Botoeva, E. (2018). Efficient han-
dling of SPARQL optional for OBDA. In The semantic Web–ISWC 2018. Proceedings,
Part I 17: 17th International Semantic Web Conference, Monterey, CA, United States,
October 8–12, 2018 (pp. 354–373). Springer International Publishing.
Federated Query Processing for Data Integration 291

Yang, F., Crainiceanu, A., Chen, Z., & Needham, D. (2023, April 1). Cluster-based joins for
federated SPARQL queries. IEEE Transactions on Knowledge and Data Engineering,
35(4), 3525–3539. https://doi.org/10.1109/TKDE.2021.3135507
Yannakis, T., Fafalios, P., & Tzitzikas, Y. (2018). Heuristics-based query reordering for feder-
ated queries in SPARQL 1.1 and sparql-ld. arXiv preprint arXiv:1810.09780.
Index
adaptive learning strategies 93 complying with regulations 240
additional results and comparison 107 components and framework overview 100
addressing challenges and future directions 129 computational efficiency 268
adopting FL AI-driven pedagogy 23 concepts and principles 90
adversarial attacks 78 consensus and validation mechanisms 101
aggregation: algorithm 87; challenges presented consensus mechanism 105
by 241; on encrypted data 88; equation 136; considerations and obstacles to overcome 240
methods that are commonly used 241 continuous growth and adaptability 128
aggregator node 101 continuous learning 125
AI in petroleum processing 136 convolutional neural networks 69
ANPR limitation factors 59 critical applications of AI in petroleum
architecture 69 processing 137
artificial general intelligence 209; challenges
and considerations in 210; key features and data analysis 25
characteristics of 210 data balancing 258
augmented reality 219 data collection and preprocessing 257
automated transportation assist 121 data exposure reduction 106
automatic maintenance 121 data heterogeneity 98
data imbalance and bias 78
background and related surveys 277 data integration and interoperability 277
background study 114 data partitioning and encryption 105
basic benefits of FL data analytics; modular data set description 106
learning 124 data storage 100
blockchain 217; layer 100; technology 94 dealing with disparate signals 129
blockchain-based authentication and identity decentralization 96; of education through AI for
management 105 the sociotechnical ecosystem 42
breaking down complex frameworks 244 decentralized learning 235
decentralized method: benefits of 230
calculation parameters 269 deciding on the best strategy 243
case studies of federated learning in petroleum decreased relations expense 129
processing 148 deep learning algorithms 69
case study: leveraging blockchain-enhanced different approaches to data source
federated learning for enhanced privacy, selection 283
security, and efficiency at TechSolve INC 105 differential privacy 160
challenges and future directions 122, 150 digital twins 214
challenges and future work 77 discussion 50
challenges and prospective courses of action 242 diversity of data 164
challenges in collaborative machine learning 96 dynamic adaptation 79
client nodes 100
CNNs are popular because of the following edge computing 202
factors 69 edge device collaborations 104
cognitive computing 217 efficient edge computing 127
collaborative efforts 79 encryption of model updates 88
collaborative innovation 144 end background study 119
collaborative model training 91, 166 enhancements to privacy and security 129
commercial UAVs 206; benefits of 207 ensuring data privacy 103
communication overlap 164 ethical and legal considerations 11
comparison with centralized approaches 107 ethical considerations 79
comparison with other learning paradigms 11 evaluation metrics 265
complexity analysis 106 experimental result 269
compliance with regulatory requirements 146 experimental setup 264

293
294Index

federated learning: advantages of using 62; ontology-based data integration for distributed
algorithms 260; applications of 9 (in intrusion data access 280
detection 245; in smart transportation 120); optimization of operational processes 143
basic concepts of 231; benefits and limitations overall perception and category of respondents’
of fundamentals of 246; challenges and responses toward sociotechnical AI 45
future of 163; classification of 232; in fake overcoming data limitations 166
currency detection 74; fundamentals of 73,
89; history of 3; in industry and research 10; perception 210
key components of 4; model in English class petroleum processing: benefits of AI in 138;
18; need for 158; taxonomy 20; types of 61; challenges of AI in 138
vertical 61; working principles of 61; works 5 privacy-preserving communication with
federated studies and analytics for data homomorphic encryption 88
throughout the IoV 122 procedural steps to do ANPR 59
federated vs traditional learning 234
financial services and banking 10 quantum computing 220
FL data process steps via application delivery query evaluation with optional in SPARQL query
network (ADN) and data analytics 122 287
future directions and emerging trends 12
rationale for integration 99
Gartner hype cycle 199 reduced data exposure 62
generalization across currency types 78 reduced latency 62
greater efficiency of transportation and regulatory and compliance requirements 99
sustainable society 130
scalability 62
healthcare and medical research 9 security concerns 97
heterogeneity handling 93 smart contracts for model aggregation 87
horizontal federated learning 161 smart sociotechnical ecosystem through
federated learning model 39
improving refinery safety with FL 149 software-defined networks: applications and
incorporating new knowledge 166 benefits of 208
industry applications 103 stochastic gradient descent with momentum 72
internet of things 168 the stages involved in a typical FL process 7
interoperability and integration 79
IoT and edge computing 104 teaching model incorporating FL 24
IoV ecosystem: challenges in 116 trade-off between privacy and utility 89
IoV paradigm 114
iterative learning rounds 92 uniformity and coordination 122
uniformity and integration 129
machine learning 213; applications of 213 user-level privacy concerns 98
method and participants 22
methodology 44 vehicle data distribution 114
models customization 122 virtual reality 204
multidisciplinary processing of data 128 virtual representation 214
volatility and ability to scale 129
natural language understanding 210
network efficiency 129 watermark and electrotype watermark 68