UNIVERSITY OF JYVÄSKYLÄ
Lecture 1: Introduction to
Cognitive Computing and Deep Learning
TIES4911 Deep-Learning for Cognitive Computing for Developers
Spring 2020
by:
Dr. Oleksiy Khriyenko
IT Faculty
University of Jyväskylä
UNIVERSITY OF JYVÄSKYLÄ
Acknowledgement
I am grateful to all the creators/owners of the images that I found from Google and have
used in this presentation.
10/01/2020 TIES4911 – Lecture 1 2
UNIVERSITY OF JYVÄSKYLÄ
IBM Deep Blue
In 1996 and 1997 years, there was a pair of six-game chess matches
between world chess champion Garry Kasparov and an IBM
supercomputer called Deep Blue.
https://en.wikipedia.org/wiki/Deep_Blue_versus_Garry_Kasparov
§ Kasparov won the first match in
Philadelphia in 1996 (4:2).
§ In 1997 in New York City, the first
computer program Deep Blue
defeated a world champion in a
match under tournament
regulations (3.5:2.5)
10/01/2020 TIES4911 – Lecture 1 3
UNIVERSITY OF JYVÄSKYLÄ
IBM Watson
Watson is an artificially intelligent cognitive computer system capable
of processing large amounts of unstructured data and answering to
queries posed in natural language.
https://www.youtube.com/watch?v=_Xcmh1LQB9I
Jeopardy! Challenge.
In February 2011, IBM Watson made history competed against the
world’s best Jeopardy! Champions (Ken Jennings and Brad Rutter).
https://www.youtube.com/watch?v=P18EdAKuC1U
10/01/2020 TIES4911 – Lecture 1 4
UNIVERSITY OF JYVÄSKYLÄ
Google's AlphaGo
Google's AlphaGo A.I. beats world's number one in Go
https://www.cnbc.com/2017/05/23/googles-alphago-a-i-beats-worlds-number-one-in-ancient-game-of-go.html
§ In October 2015, the distributed
version of AlphaGo defeated the
European Go champion Fan Hui
(2-dan)(5:0)
§ In March 2016, in a five-game
match, the first time a computer
Go program has beaten a 9-dan
professional Lee Sedol (one of the
best players)(4:1)
§ In May 2017, in a three-game
match, Global Champ Chinese
player Ke Jie(9-dan), lost his first
game against Google DeepMind
computer program AlphaGo.
AlphaGo Zero
Introduces in October 2017, the latest evolution of AlphaGo is even more powerful and is
arguably the strongest Go player in history.
https://deepmind.com/blog/alphago-zero-learning-scratch
Minigo: an open-source implementation of the AlphaGoZero algorithm (https://github.com/tensorflow/minigo)
In December 2017, DeepMind claimed that it generalized AlphaGo Zero's approach into a single AlphaZero
algorithm. https://en.wikipedia.org/wiki/AlphaZero
Mastering the game of Go without human knowledge
https://www.nature.com/articles/nature24270.epdf
10/01/2020 TIES4911 – Lecture 1 5
UNIVERSITY OF JYVÄSKYLÄ
Intelligent Robots
An Artificial Super Intelligent Robot Sophia wants
to study at collage, wants job, family, citizenship, etc.
Sophia is a social humanoid robot developed by Hong Kong-
based company Hanson Robotics in 2015. In October 2017,
the robot became a Saudi Arabian citizen, the first robot to
receive citizenship of any country.
§ https://en.wikipedia.org/wiki/Sophia_(robot)
§ http://dailyinformator.com/8-things-you-need-to-know-about-sophia-worlds-
first-robot-citizen/
§ https://www.youtube.com/watch?v=IUJaO6C-zTo
§ https://www.youtube.com/watch?v=8MjIU4eq__A
§ https://www.youtube.com/watch?v=fLvL7uqrMVc
§ https://www.youtube.com/watch?v=S5t6K9iwcdw
§ https://www.youtube.com/watch?v=XwRXv80AUTs
§ …
10/01/2020 TIES4911 – Lecture 1 6
UNIVERSITY OF JYVÄSKYLÄ
Intelligent Robots
Microsoft and Alibaba AI programs beat humans in Stanford reading
comprehension test for 1st time
Machines can already outplay us in chess, poker and other
games, and now they are becoming better readers as well.
AI programs from both Microsoft and Alibaba outperformed
humans in the beginning of January 2018 on a reading
comprehension data set developed at Stanford.
“Crowdworkers” scraped more than 500 Wikipedia articles to
produce more than 100,000 question-and-answer sets for the
test.
Here’s a sample question: “What year did Genghis Khan die?”
(Spoiler alert: It’s 1227.)
“This is the first time that a machine has outperformed humans
on such a test,” Alibaba said in a statement.
Microsoft’s score of 82.6 and Alibaba’s grade of 82.4 beat out
the human standard of 82.3. Other notable AI programs
participating in the test and closing in on beating human scores
come from the Allen Institute for Artificial Intelligence, Tencent,
Salesforce and others.
§ https://www.geekwire.com/2018/microsoft-alibaba-ai-programs-beat-
humans-stanford-reading-test-1st-time/
10/01/2020 TIES4911 – Lecture 1 7
UNIVERSITY OF JYVÄSKYLÄ
Intelligent Robots
Intelligent Virtual Assistants and Chat Bots…
Amazon Alexa Google Home Apple Siri
https://en.wikipedia.org/wiki/Amazon_Alexa https://assistant.google.com/ https://www.apple.com/siri/
IBM Watson Assistant Microsoft Cortana Facebook Chatbot
https://www.ibm.com/cloud/watson-assistant/ https://www.microsoft.com/en-us/cortana https://www.microsoft.com/en-us/cortana
10/01/2020 TIES4911 – Lecture 1 8
UNIVERSITY OF JYVÄSKYLÄ
Self-Driving Cars
NVIDIA and Bosch Announce AI Self-Driving Car Computer
https://blogs.nvidia.com/blog/2017/03/16/bosch/
10/01/2020 TIES4911 – Lecture 1 9
UNIVERSITY OF JYVÄSKYLÄ
Intelligent Robots
BOSTON DYNAMICS is a world leader in mobile robots, tackling some of the toughest
robotics challenges. https://www.bostondynamics.com
It combines the principles of dynamic control and balance with
sophisticated mechanical designs, cutting-edge electronics, and next-
generation software for high-performance robots equipped with
perception, navigation, and intelligence. Boston Dynamics has an
extraordinary and fast-growing technical team of engineers and
scientists who seamlessly combine advanced analytical thinking with
bold engineering and boots-in-the-mud practicality.
§ https://www.youtube.com/user/BostonDynamics
§ https://www.youtube.com/watch?v=NR32ULxbjYc
10/01/2020 TIES4911 – Lecture 1 10
UNIVERSITY OF JYVÄSKYLÄ
Autonomous Delivery
Robot delivery dogs deployed by self-driving cars …
https://techcrunch.com/2019/01/07/robot-delivery-dogs-deployed-by-self-driving-cars-are-coming/
Let’s hope you’re not
afraid of dogs, because if
Continental gets its way,
autonomous robot dogs
are going to be delivering
your packages.
10/01/2020 TIES4911 – Lecture 1 11
UNIVERSITY OF JYVÄSKYLÄ
Autonomous Delivery
Agility’s two-legged robot Digit is for sale and Ford is
the first customer…
https://techcrunch.com/2020/01/05/agilitys-two-legged-robot-digit-is-for-sale-and-ford-is-the-first-customer/
10/01/2020 TIES4911 – Lecture 1 12
UNIVERSITY OF JYVÄSKYLÄ
Intelligent Robots
Inventory Robots in Retailing…
§ https://www.solteq.com/en/solutions/solteq-retail-robot?
§ https://www.youtube.com/watch?v=nda7bTLNcQs
10/01/2020 TIES4911 – Lecture 1 13
UNIVERSITY OF JYVÄSKYLÄ
Intelligent Robots
Robots in Education…
10/01/2020 TIES4911 – Lecture 1 14
UNIVERSITY OF JYVÄSKYLÄ
Intelligent Robots
Robots in Service…
10/01/2020 TIES4911 – Lecture 1 15
UNIVERSITY OF JYVÄSKYLÄ
Neuromorphic Computing
Neuromorphic Computing: The Future of AI and Computing
Family of Xeon® Scalable Processors and Intel®
Intel®
Nervana™ Neural Network Processors (NNP) Redefine AI Silicon
§ https://www.intel.com/content/www/us/en/benchmarks/server/xeon-scalable/xeon-
scalable-artificial-intelligence.html
§ https://software.intel.com/en-us/articles/intel-processors-for-deep-learning-training
§ https://ai.intel.com/intel-nervana-neural-network-processors-nnp-redefine-ai-
silicon/?utm_source=ISTV&utm_medium=Video&utm_campaign=ISTV2018_ISTV18
02_01&utm_content=AI+News
§ https://www.youtube.com/watch?v=zEzm-rMwyVo
§ https://www.youtube.com/watch?v=ej9-sGj_iHw
Intel Loihi AI chip A new AI chip under development at Intel
is taking its inspiration from the human brain in an attempt to
overcome technological hurdles. This New Self-Learning Chip
Promises to Accelerate Artificial Intelligence
§ https://www.youtube.com/watch?v=EgCRwZw4p8c
§ https://www.youtube.com/watch?v=NAE4YUB09j4
§ https://newsroom.intel.com/editorials/intels-new-self-learning-chip-promises-
accelerate-artificial-intelligence/
10/01/2020 TIES4911 – Lecture 1 16
UNIVERSITY OF JYVÄSKYLÄ
?!
“Slaughterbots” is seven minutes and forty-seven seconds of sheer horror
designed to be a fictional warning against a future full of killer robots.
This fictional video about AI-powered weapons makes The Terminator look like a Disney film
§ https://thenextweb.com/artificial-intelligence/2017/11/14/this-fictional-video-about-ai-powered-weapons-makes-the-terminator-look-like-a-
disney-film/?utm_content=buffer215b3&utm_medium=social&utm_source=facebook.com&utm_campaign=buffer
§ https://www.youtube.com/watch?v=9CO6M2HsoIA
§ http://autonomousweapons.org
10/01/2020 TIES4911 – Lecture 1 17
UNIVERSITY OF JYVÄSKYLÄ
?!
Something went wrong…
https://www.youtube.com/watch?v=y3RIHnK0_NE
https://www.youtube.com/watch?v=ZoemTySxFso
10/01/2020 TIES4911 – Lecture 1 18
UNIVERSITY OF JYVÄSKYLÄ
?!
Deep Fake…
https://www.youtube.com/watch?v=78L6I6vsfrU
10/01/2020 TIES4911 – Lecture 1 19
UNIVERSITY OF JYVÄSKYLÄ
Cognitive Computing
Cognitive Computing is a new type of
computing with the goal of more accurate
models of how the human brain/mind
senses, reasons, and responds to stimulus.
https://en.wikipedia.org/wiki/Cognitive_computing
Cognitive Computing based systems are “systems that learn at scale,
reason with purpose and interact with humans naturally.” (IBM)
Types of cognitive technologies:
q Machine learning
q Natural language processing
q Speech recognition
q Computer vision
q Insights generation from data
q Sentiment analysis
q Etc.
10/01/2020 TIES4911 – Lecture 1 20
UNIVERSITY OF JYVÄSKYLÄ
Cognitive Computing
On the way towards Cognitive Computing, smart
systems adopt key elements of cognitive computing…
q Expanding the boundaries of human cognition
… extends a capability of a human to reason, think deeply,
recognize objects and sounds, manipulate and manage huge
amount of data (not only to search in big volume, but make
decisions on top of it).
q More natural human-computer interaction
… applies more natural interaction and engagement with
computers via more general speech and natural language
communication with the system, as well as, use of infographics
and visual data representation techniques.
q Use of Learning
… helps to design personalized and adaptable systems able to
constantly learn and evolve based on feedback from used
interaction applying machine learning, statistics, etc.
10/01/2020 TIES4911 – Lecture 1 21
UNIVERSITY OF JYVÄSKYLÄ
Deep Learning
Machine Learning - process of training a machine to create a model and use it for
decision making.
Deep Learning - is part of a broader family
of machine learning methods based on learning
data representations, as opposed to task-
specific algorithms.(Wikipedia)
“The analogy to deep learning is that the rocket engine is the deep
learning models and the fuel is the huge amounts of data we can
feed to these algorithms.” (Andrew Ng)
Deep Learning vs. Machine Learning – the essential differences you need to know!
https://www.analyticsvidhya.com/blog/2017/04/comparison-between-deep-learning-machine-learning/
https://www.zendesk.com/blog/machine-learning-and-deep-learning/
https://www.edureka.co/blog/deep-learning-tutorial
https://www.youtube.com/watch?v=vehXkgG3YcU
10/01/2020 TIES4911 – Lecture 1 22
UNIVERSITY OF JYVÄSKYLÄ
Deep Learning
DL use-cases
q Computer/Machine Vision
§ image classification and automatic tagging
§ object recognition in the image
§ video recognition
q Speech recognition
and generation
q Text processing
§ fact extraction
§ machine translation
§ sentiment analysis
§ character level text processing
§ document classification
q Decision making
q Etc.
10/01/2020 TIES4911 – Lecture 1 23
UNIVERSITY OF JYVÄSKYLÄ
Deep Learning
DL application domains
q Medical
§ cancer detection
§ drug discovery
§ radiology (CNN based detection of tumor and cancer
from MRI, fMRI, EKG, and CT scans)
q Finance
§ market, trading and investment predictions
§ customer segmentation in advertising
§ fraud detection
q Agriculture
§ problematic environmental conditions detection based
on satellite feeds and sensor data
q Smart Cities
q Traffic &Transportation
q Gaming
q Music and Art
q Robotics
q Education
q Service domains
q Etc.
10/01/2020 TIES4911 – Lecture 1 24
UNIVERSITY OF JYVÄSKYLÄ
Deep Learning
DL application domains
q Forecasting Waves with Deep Learning
https://www.engineering.com/DesignSoftware/DesignSoftwareArticles/A
rticleID/15888/Forecasting-Waves-with-Deep-Learning.aspx
q CheXNet: Radiologist-Level Pneumonia
Detection on Chest X-Rays with Deep
Learning
§ https://stanfordmlgroup.github.io/projects/chexnet/
§ http://med.stanford.edu/news/all-news/2017/11/algorithm-can-
diagnose-pneumonia-better-than-radiologists.html
q Face It – The Artificially Intelligent
Hairstylist
https://software.intel.com/en-us/articles/face-it-the-artificially-intelligent-
hairstylist?utm_source=taboola&utm_medium=referral&utm_campaign=
AI_Student_ASMO_Q4_2017_Media_Buy
10/01/2020 TIES4911 – Lecture 1 25
UNIVERSITY OF JYVÄSKYLÄ
Deep Learning
DL most known researchers
q Andrew Ng
q Geoff Hinton
q Yann LeCun
q Yoshua Bengio
q Andrej Karpathy
q …
Comprehensive Neural Network
learning materials
q Michael Nielsen's book:
http://neuralnetworksanddeeplearning.com/
q Andrew Ng's classes:
http://www.andrewng.org/courses/
https://www.coursera.org/learn/machine-learning
http://cs229.stanford.edu/
https://www.deeplearning.ai/
10/01/2020 TIES4911 – Lecture 1 26
UNIVERSITY OF JYVÄSKYLÄ
Deep Learning
DL companies (big players) BOSCH
supports development of autonomous vehicles
Apple
Google is actively investing into self-driving cars
bought DeepMind for $ 400 Million NVIDIA
and provides Cognitive Computing services improved DL hardware with GPUs
Intel
introduced Intel Nervana NN processor and IBM Watson
new Loihi Self-Learning Chip, and improves Cognitive Computing services Facebook AI
DL toolkits, frameworks and algorithms… DL frameworks and tools
Microsoft Azure
Amazon Cognitive services and toolkit
AWS Cloud Cognitive Services Toyota
has invested $ 1 Billion into AI research
10/01/2020 TIES4911 – Lecture 1 27
UNIVERSITY OF JYVÄSKYLÄ
Deep Learning Frameworks
Intel
Relevant links:
https://towardsdatascience.com/top-10-best-deep-learning-frameworks-in-2019-5ccb90ea6de
https://towardsdatascience.com/battle-of-the-deep-learning-frameworks-part-i-cff0e3841750
https://towardsdatascience.com/deep-learning-framework-power-scores-2018-23607ddf297a
https://www.marutitech.com/top-8-deep-learning-frameworks/
https://www.netguru.com/blog/deep-learning-frameworks-comparison
10/01/2020 TIES4911 – Lecture 1 28
UNIVERSITY OF JYVÄSKYLÄ
Deep Learning Frameworks
q TensorFlow by Google seems to be the most used deep learning framework so far. In 2017, Google has
introduced TensorFlow Lite - is a lightweight solution for mobile and embedded devices, enabling on-device machine
learning inference and supporting hardware acceleration with the Android Neural Networks API.
https://www.tensorflow.org/
q PyTorch was introduced by Facebook in January 2017 and already started to gain popularity. The main drivers
behind the popularity are GPU acceleration, the efficient usages of memory and the use of dynamic computational
graphs (“define by run” instead of the traditional “define and run”). http://pytorch.org/
q Caffe2 framework has been launched by Facebook in 2017 as the successor of the well known and still extremely
popular Caffe framework. https://caffe2.ai/ and http://caffe.berkeleyvision.org/
q MXNet is Apache library for deep learning supported by Microsoft and Amazon. It supports many languages, from
C++ to Python, JavaScript, Go, and, indeed, R. https://mxnet.apache.org/
q CNTK deep learning framework developed by Microsoft. The framework was renamed to the Microsoft Cognitive
Toolkit with officially launched the 2.0 version in 2017. https://www.microsoft.com/en-us/cognitive-toolkit/
q Torch is a scientific computing framework for LuaJIT with wide support for machine learning algorithms that puts
GPUs first. http://torch.ch/
q Theano is a Python library that allows to efficiently define, optimize, and evaluate mathematical expressions
involving multi-dimensional arrays. http://deeplearning.net/software/theano/
q Deeplearning4j is an open-Source, distributed, deep Learning Library for the JVM. https://deeplearning4j.org/
q Chainer is a Python-based deep learning framework aiming at flexibility. IntelChainer is Optimized-Chainer
for Intel Architectures. https://chainer.org/, https://github.com/chainer/chainer and https://github.com/intel/chainer
q Neon is a deep learning (Python-based and optimized for Intel® architecture) framework designed for ease of use and
extensibility on modern deep neural networks, such as AlexNet, Visual Geometry Group (VGG), and GoogLeNet.
https://software.intel.com/en-us/ai-academy/frameworks/neon/ and https://ai.intel.com/neon/
Intel provides Framework Optimizations for faster training of deep
neural networks on Intel architecture https://ai.intel.com/framework-optimizations/
10/01/2020 TIES4911 – Lecture 1 29
UNIVERSITY OF JYVÄSKYLÄ
Deep Learning Frameworks
Interfaces that are wrapped around one or multiple frameworks:
q Keras is the most well know and widely used interface for deep learning. This high-level Python based deep
learning API is created by a deep learning researcher at Google - François Chollet. Google announced in 2017 that
Keras has been chosen to serve as the high-level API of TensorFlow and will be included in the next TensorFlow
release. Next to TensorFlow, Keras can also use Theano or CNTK as backend. https://keras.io/
q Gluon is an open source high-level Python deep learning interface which allows developers to more easily and
quickly build machine learning models, without compromising performance was jointly announced by Microsoft and
Amazon’s AWS in October 2017. Interface wraps MXNet and soon it will also include Microsoft’s CNTK.
https://mxnet.incubator.apache.org/gluon/index.html
q Eager execution for TensorFlow, introduced in October 2017, is an imperative “define-by-run” interface
where operations are executed immediately as they are called from Python. With this launch, Google hopes to win
back the users that fell in love with PyTorch and it’s dynamic graph.
https://research.googleblog.com/2017/10/eager-execution-imperative-define-by.html,
https://github.com/tensorflow/tensorflow/blob/master/tensorflow/contrib/eager/python/g3doc/guide.md
q Sonnet is a library (by DeepMind) built on top of TensorFlow for building complex neural networks.
https://deepmind.com/blog/open-sourcing-sonnet/
q H2O Deep Water is a H2O for GPU Enabled Deep Learning on all data types integrating with TensorFlow,
MXNet and Caffe. Deep Water brings all these frameworks together under the same user interfaces as the H2O
platform. Now, in addition to the original H2O Deep Learning algorithm, users can access TensorFlow, MXNet and
Caffe backends in H2O, and build complex deep networks. https://www.h2o.ai/deep-water/
ONNX (Open Neural Network Exchange) (is launched by Microsoft, AWS, and Facebook amongst others)is an
open format to represent deep learning models that allows users to more easily move models between different frameworks (V1 is
released in December 2017). ONNX enables models to be trained in one framework and transferred to another for inference. ONNX
models are currently supported in Caffe2, Microsoft Cognitive Toolkit, MXNet, and PyTorch, and there are connectors for many other
common frameworks and libraries (community already added a converter for TensorFlow as well). https://onnx.ai/ and
https://github.com/onnx/onnx
10/01/2020 TIES4911 – Lecture 1 30
UNIVERSITY OF JYVÄSKYLÄ
Deep Learning Platforms
q H2O.ai is a machine learning platform. https://www.h2o.ai/
q Spark is a fast and general engine for large-scale data processing. http://spark.apache.org/
q PlaidML - open source portable deep learning engine from Vertex.AI.
http://vertex.ai/blog/announcing-plaidml
https://github.com/plaidml/plaidml
q A.I. Model - a common machine learning tool for all frameworks by Microsoft.
https://www.microsoft.com/en-us/AI/ai-platform
three major tools for developers:
§ Azure Machine Learning Experimentation Service
§ Azure Machine Learning Workbench
§ Azure Machine Learning Model Management Service.
q IBM Watson and corresponding Watson Data Platform and IBM Services Platform.
https://www.ibm.com/watson/
q Google Cloud AI provides modern machine learning services, with pre-trained
models and a service to generate your own tailored models.
https://cloud.google.com/products/machine-learning/
q Intel® AI DevCloud for Intel® AI Academy members.
https://software.intel.com/ai-academy/tools/devcloud
10/01/2020 TIES4911 – Lecture 1 31
UNIVERSITY OF JYVÄSKYLÄ
Cognitive Computing Services
Google (DeepMind)
https://cloud.google.com/vision/ IBM Watson
https://cloud.google.com/speech/ https://www.ibm.com/watson/services/visual-recognition/
… https://www.ibm.com/watson/services/text-to-speech/
https://www.ibm.com/watson/services/natural-language-understanding/
…
Microsoft Azure
https://azure.microsoft.com/en-us/services/cognitive-services/computer-vision/ Amazon
https://azure.microsoft.com/en-gb/services/cognitive-services/speech/ https://aws.amazon.com/rekognition/
https://azure.microsoft.com/en-us/services/cognitive-services/face/
https://azure.microsoft.com/en-us/services/cognitive-services/text-analytics/
…
https://www.captionbot.ai/
https://www.how-old.net/
https://www.what-dog.net/ Facebook
https://ai.facebook.com/tools/
Clarifai
https://www.clarifai.com/
Cognitec MetaMind
http://www.cognitec.com/ Kairos https://einstein.ai/
https://www.kairos.com/
https://metamind.readme.io/
Relevant links:
https://blog.filestack.com/thoughts-and-knowledge/comparing-google-vision-microsoft-cognitive-amazon-rekognition-clarifai/
10/01/2020 TIES4911 – Lecture 1 32
