A Causal Introduction to
Machine Learning
organized by
GCT IEEE Student Branch
Government College of Technology, Coimbatore
October 19, 2024
Mukil Saravanan
© 2024 Mukil Saravanan
Prompt: A movie trailer featuring the adventures of the 30 year old
space man wearing a red wool knitted motorcycle helmet, blue sky,
salt desert, cinematic style, shot on 35mm film, vivid colors.
Credits: Sora - OpenAI © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 1
Prompt: An astronaut riding a horse on the moon
Credits: Free AI Image Generator - Image Creator in Bing © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 2
Credits: Reaching the Limit in Autonomous Racing: Optimal Control versus Reinforcement Learning (SciRob 23) © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 3
Credits: Godfather of AI who Opposes Autonomous Weapons Wins Nobel Prize in Physics - Mines Action Canada © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 4
Session goals
➢ This session will make you understand the overview of Machine
Learning
➢ It is designed to make you aware of different state-of-the-art
Machine Learning models
➢ It will provide you with necessary motivation to get started with
Machine Learning
© 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 5
Time Breakdown
➢ [10 mins] – Motivation, Applications, Learning objectives of the
session
➢ [20 mins] – Machine Learning and its types
➢ [20 mins] – Deep Learning frameworks, State-of-the-art models
➢ [5 mins] – Q&A
➢ [5 mins] – Feedback
© 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 6
Let’s get into it!
© 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 7
Why AI?
➢ Unsolvable by pre-determined rules
➢ Good use of existing data
➢ To understand human learning
© 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 8
Credits: The AI Conversation - AI, ML, DL & LLMs (youtube.com) © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 9
Types of Learning
Credits: An Overview of Supervised, Unsupervised, and Reinforcement Learning in Machine Learning (c-sharpcorner.com) © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 10
Supervised Learning
Credits: Types of Machine Learning - Supervised, Unsupervised, Reinforcement - TechVidvan © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 11
Supervised Learning – Types
Credits: Classification and Regression in Machine Learning (enjoyalgorithms.com) © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 12
Unsupervised Learning - Clustering
Credits: Types of Machine Learning - Supervised, Unsupervised, Reinforcement - TechVidvan © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 13
Unsupervised Learning – Dimensionality Reduction
3D manifold → 2D space
Credits: 14_Dimensionality_Reduction (holehouse.org) © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 14
Reinforcement Learning
Credits: Reinforcement learning on a chicken (youtube.com) © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 15
Reinforcement Learning
Credits: Types of Machine Learning - Supervised, Unsupervised, Reinforcement - TechVidvan © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 16
Linear Regression
➢ Use case: Linear data →
Predict in a continuous
range
➢ Example: Finding weight
of a person with height
data
Credits: Lineaire regressie in machinaal leren - GeeksforGeeks © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 17
Logistic Regression = Binary Classification
➢ Use case: Linear data →
classify two different
classes
➢ Example: Finding obese
or not obese with weight
data
Credits: Logistic Regression in Machine Learning | Wireless Pi © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 18
Why Deep Learning?
© 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 19
Features extraction is crucial for learning
Credits: Demystifying Deep Learning: The Future of AI | by Hassaan Idrees | Medium © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 20
Manual Feature Extraction
Credits: MIT 6.S191: Convolutional Neural Networks - YouTube © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 21
Biological Neuron vs Artificial Neuron
Credits: The differences between Artificial and Biological Neural Networks | by Richard Nagyfi | Towards Data Science © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 22
Deep Learning
Credits: MIT 6.S191: Convolutional Neural Networks - YouTube © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 23
Common Activation Functions
All activation functions
are non-linear
Credits: MIT 6.S191: Convolutional Neural Networks - YouTube © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 24
Non-linearity
Credits: MIT 6.S191: Convolutional Neural Networks - YouTube © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 25
Single Layer Neural Network
Credits: MIT 6.S191: Convolutional Neural Networks - YouTube © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 26
Loss Function
Credits: MIT 6.S191: Convolutional Neural Networks - YouTube © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 27
Loss Optimization
Credits: MIT 6.S191: Convolutional Neural Networks - YouTube © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 28
Gradient Descent
Credits: MIT 6.S191: Convolutional Neural Networks - YouTube © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 29
OK, let’s take a deep
breath
© 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 30
Fully Connected Network (FCN)
Credits: 6S191_MIT_DeepLearning_L3.pdf (introtodeeplearning.com) © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 31
Convolutional Neural Network (CNN)
➢ Bottleneck: FCN can’t
handle images very well
➢ Core Idea: Finds
similarity & learns
hierarchal features
➢ Position Invariant
Credits: 6S191_MIT_DeepLearning_L3.pdf (introtodeeplearning.com) © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 32
Convolutional Neural Network (CNN)
Credits: 6S191_MIT_DeepLearning_L3.pdf (introtodeeplearning.com) © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 33
Transfer Learning
Credits: The architecture of our transfer learning model. | Download Scientific Diagram (researchgate.net) © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 34
CNN Applications
Credits: Object Detection vs. Classification in Computer Vision: Explained (augmentedstartups.com) © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 35
CNN Applications
Credits: Variational End-to-End Navigation and Localization (youtube.com) © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 36
Sequence Modelling
Audio, sentences, biological signals and so on are sequences
Credits: 6S191_MIT_DeepLearning_L2.pdf (introtodeeplearning.com) © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 37
Recurrent Neural Network (RNN)
Credits: 6S191_MIT_DeepLearning_L2.pdf (introtodeeplearning.com) © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 38
RNN Applications
Credits: 6S191_MIT_DeepLearning_L2.pdf (introtodeeplearning.com) © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 39
Generative AI Models
Diffusion Models Large Language Models
Credits: How to Get Better Outputs from Your Large Language Model | NVIDIA Technical Blog © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 40
Foundation Models
Images Videos Robots
Credits: DALL·E 3 | OpenAI, Sora | OpenAI, RT-2: Vision-Language-Action Models (robotics-transformer2.github.io) © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 41
Summary
➢ Different layers of AI
➢ Types of learning – supervised, unsupervised, reinforcement
➢ Linear, Logistic regression
➢ Perceptron model, Multi-Layer Perceptron network
➢ Loss function, Gradient Descent
➢ Various DL models: CNN, RNN, generative models, foundation
models
© 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 42
Q&A
Credits: Albert Einstein Quote: “The important thing is not to stop questioning. Curiosity has its own reason for existing.” (quotefancy.com) © 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 42
© 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 43
Please help me optimize my loss
function (pun intended)
© 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 44
Thank You
To know about my works, visit
https://mukilsaravanan.github.io/
Reach out to me
Email: mukil.saravanan.edu@gmail.com
LinkedIn: Mukil Saravanan
© 2024 Mukil Saravanan
19-OCT-2024 A CAUSAL INTRODUCTION TO MACHINE LEARNING 45
