Introduction to Machine Learning

Lecture 1
Date : 2 Jan, 2025

Apurva Dhingra
PhD @ CMInDS, IIT Bombay
Lecture -1 Outline
Introduction to Machine Learning:
● Definition of machine learning.
● Distinction between traditional programming and machine learning.
Historical Context:
● Milestones in the development of machine learning.
● Key breakthroughs and their impact.
Applications in Physics:
● Examples of successful machine learning applications in physics.
● Highlighting the potential for solving complex problems in physics using machine learning.
Types of Machine Learning:
● Brief introduction to supervised learning, unsupervised learning, and reinforcement learning.
● High-level explanation of how each type works.
Overview of the Course Structure:
● Outline of topics covered in subsequent weeks.
● Emphasis on the balance between theory and practical implementation.
Lecture-1 outline continued
Identifying Problems Suitable for ML:
● Discussion on the types of physics problems that can benefit from machine learning.
● Examples of problems where data-driven approaches can enhance traditional methods.
Overview of Physics Datasets:
● Introduction to common datasets used in physics research.
● Discussion on the challenges and opportunities in handling physics data.
Introduction to Python:
● Basics of Python programming for data science.
● Overview of Python's role in machine learning.
Libraries for Data Handling:
● Introduction to NumPy and Pandas for efficient data manipulation.
● Demonstrations of basic data preprocessing techniques.
Content

Introduction to ML

ML vs Traditional programming

Applications of ML

Components of ML

Types of ML

Physics and ML
Introduction : AI and ML
AI : Artificial intelligence (AI), the ability of a digital computer or computer-controlled robot to perform tasks commonly
associated with intelligent beings.

Ex - rule-based systems like chatbots

ML : Machine Learning is a field of artificial intelligence that focuses on the development of algorithms and statistical models
allowing computer systems to perform tasks without explicit programming.

Ex - ChatGPT
Machine Learning

● Ability of computers to “learn” from “data” or “past experience”.

● data: Comes from various sources such as sensors, domain knowledge,
experimental runs, etc.
● learn: Make intelligent predictions or decisions based on data by optimizing a
model

“All models are wrong, but some are useful”,

George E. P. Box
Traditional Programming vs. Machine Learning

● Traditional Programming ● Machine Learning

● Traditional Programming: Rules are explicitly ● Machine Learning: Algorithms learn patterns and
programmed to solve a specific task. rules from data to perform tasks without explicit
● Traditional Programming Example: Sorting programming.
algorithms, if-else statements. ● Machine Learning Example: Predicting house
prices, recognizing handwritten digits.
Characteristics of Machine Learning

● Adaptability: ML systems can learn and adapt from experience.

● Automation: ML automates the process of gaining insights and
making decisions.
● Iterative Improvement: ML models improve over time as they are
exposed to more data.
When do we use ML?
For tasks that are easily performed by humans but are complex for computer
systems to emulate
Find the
squares
with traffic
lights

Sheepdog
or mop
Applications of Machine Learning

● Image and Speech Recognition

● Natural Language Processing
● Recommendation Systems
● Autonomous Vehicles
 Key Components of Machine Learning

● Features: Input variables used by the model for predictions/classification-

the x variable in simple linear regression(curve fitting).
● Labels: Output or target variable in supervised learning - the y variable in simple
linear regression.
● Model: The algorithm or mathematical structure used for learning - relation
between feature and labels (in supervised learning)

Example - Spam detector

Features : words in the email text , sender's address, etc

Label : individual emails that users have explicitly marked as "spam" or "not spam."
Components of Machine learning
 Workflow of Machine Learning MODEL DATA
● Data Collection: Gathering relevant
data for the problem at hand.
● Data Preprocessing: Cleaning,
transforming, and preparing data for
analysis.
● Model Training: Training the model
on the training dataset.
● Evaluation: Assessing the model's
performance on a separate dataset.
● Prediction: Using the trained model
to make predictions on new, unseen
data.

Optimisation Algorithms
How do we find the best setting of
switches to match the data?
Types of Machine Learning

Type of ML Different methods used for this What is it?

● Supervised Learning Linear regression, Logistic ● Learning from labeled

regression, decision trees, training data.
Neural networks

● Unsupervised Learning K-means clustering ● Extracting patterns from

unlabeled data.

● Reinforcement Learning Not covered in this course ● Learning through interaction

with an environment and
receiving feedback.
Supervised Learning (Introduction)
Regression vs. classification

A regression(prediction) model predicts continuous values. For example, regression models make predictions that answer
questions like the following:

● What is the value of a house in California?

● What is the probability that a user will click on this ad?

A classification model predicts discrete values. For example, classification models make predictions that answer questions
like the following:

● Is a given email message spam or not spam?

● Is this an image of a dog, a cat, or a hamster?
Example of Supervised Learning
Demonstration

Online demonstration : LINK

Kaggle code : LINK

Numpy Tutorial : LINK

Unsupervised learning
Activity : Applications of ML

Steps :

1. Identify a real-world application of Machine learning.

2. For the given application identify what is the Input and Output of the
application.

Is it prediction, classification, or clustering?

3. Can this task be performed without using machine learning? Can it be done by
traditional programming and domain/expert knowledge?
Application of ML - Face Recognition

Image source
Some examples from real life
Google Search Information
Retrieval
Search within articles, e-books, etc.

Face detection in photos Computer Vision

Detecting cars on roads (e.g., for self-driving cars)

Facebook recommending friends/ads Recommender

Systems
Netflix/Amazon/Flipkart recommendations

Game playing (chess, Go, poker, etc.) Robotics ML in

Expert Systems
Robots (playing soccer, robotic arm, etc.), IBM’s Watson (Jeopardy, etc.)

Medical Diagnosis Systems Speech & Natural

Language Processing
Personal Assistant (Siri, Google Assistant, Amazon Alexa, etc.)

Closed Captioning (Google Meet, MS Teams, Zoom, etc.)

History of ML
 Big Data Era (2010s)
History of ML ● Explosion of Data: Increased availability of large datasets.
● Deep Learning Revolution: Breakthroughs in deep neural
networks.
Early Years (1950s - 1960s) ● TensorFlow and PyTorch: Open-source frameworks for deep
● Dartmouth Workshop (1956): The birth of AI and the learning.
term "machine learning."
● Rosenblatt's Perceptron (1957): Initial work on Recent Advancements (2020s)
artificial neural networks.
● Transfer Learning: Leveraging pre-trained models for various
● AI Winter (1970s - 1980s)
tasks.
● Funding Cuts: Reduced interest and funding for AI
● Explainable AI: Focus on making machine learning models more
research.
interpretable.
● Expert Systems Dominance: Rule-based systems ● Quantum Machine Learning: Exploring the intersection of
dominated AI applications. quantum computing and ML.
● Renaissance (1990s)
● Emergence of Support Vector Machines (SVM):
Advances in classification algorithms. Key Breakthroughs and Their Impact
● Introduction of Decision Trees: Widely used for ● Title: Transformative Breakthroughs
classification and regression. ● Impact of AlphaGo (2016): Reinforcement learning triumphs in
game playing.
Rise of Neural Networks (2000s) ● GPT-3 (2020): Massive language model showcasing natural
language processing capabilities.
● Backpropagation Algorithm: Improvements in training
neural networks.
Continuous Evolution
● Application in Image Recognition: Success stories in
computer vision.
Physics and ML
(Statistical) Physics to ML ML to Physics

● Random Forest ● Machine learning algorithms are heavily

used to process satellite data in
● Diffusion Model - used in Generative atmospheric physics as well as handle
models weather forecasts and predictions.
● Boltzmann Machine(BM) ● Machine learning algorithms used to check
for anomalous data and to present a
summarised report for LHC data collected
at CERN
● Astrophysicists use machine learning
techniques to classify and organise data.
Datasets

Kaggle: https://www.kaggle.com/datasets
Another good resource: http://deeplearning.net/datasets/
Popular resource for ML beginners: http://archive.ics.uci.edu/ml/index.php
Interesting datasets for computational journalists:
http://cjlab.stanford.edu/2015/09/30/lab-launch-and-data-sets/
Speech and language resources: www.openslr.org/
… and so do ML libraries/toolkits
scikit-learn, openCV, Keras, Tensorflow, NLTK, etc.
References/Books

● Understanding Machine Learning (Shai Shalev-Shwartz and Shai Ben-David. Cambridge

University Press. 2017)
● The Elements of Statistical Learning (Trevor Hastie, Robert Tibshirani and Jerome Friedman.
Second Edition. 2009)
● Pattern Recognition and Machine Learning (Christopher Bishop. Springer. 2006)
● Mathematics for Machine Learning (Marc Peter Deisenroth, A. Aldo Faisal, Cheng Soon Ong.
2020)
 Thank You
QR code for the slides