Introduction to Machine Learning
• A Beginner’s Journey into Teaching Computers
to Learn
The Old Way – Traditional
Programs
• Computers used to follow fixed rules.
• Example: A calculator always knows '2+2=4'.
• But it cannot learn new tricks.
• Limitation: If rules are unknown, we can’t
program them.
The Need for Learning
• Some problems have no clear rules:
• - Recognizing a face
• - Driving a car
• - Understanding speech
• Humans do these effortlessly.
• ML lets computers find patterns instead of us
writing rules.
Data Is the New Treasure
• Every action creates data (shopping, browsing,
GPS).
• Data contains patterns about behavior.
• ML acts like a gold miner: finds precious rules
in raw data.
What Is Machine Learning?
• Programming computers to improve using
data.
• Start with a general model (many dials).
• Learning = turning the dials until model
matches data.
• Result: specialized program that can predict or
explain.
Traditional Program vs ML System
• Traditional Program (Calculator): Fixed rules,
always correct, cannot adapt.
• ML System (Spam Filter): Learns from data,
improves with time, adapts (sometimes
wrong).
Applications of ML
• Retail: predict purchases
• Finance: credit scoring, fraud detection
• Medicine: diagnosis
• Telecom: call quality
• Science: analyze big data
• Daily Life: Netflix, Amazon, Face Unlock
Association Rules
• Supermarket shopping patterns.
• Example: Beer → Chips (70%).
• Used in recommendations (books, movies,
products).
Supervised Learning
• Learning with a teacher.
• Input (X) → Output (Y).
• Model learns mapping X→Y.
• Types: Classification (categories), Regression
(numbers).
Classification
• Goal: Assign input to a category.
• Example: Loan Approval → Low-risk vs High-
risk.
• Rule: IF income > θ1 AND savings > θ2 → Low-
risk ELSE High-risk.
• Applications: Face, handwriting, speech,
medical, biometrics.
Regression
• Goal: Predict a number.
• Example: Used car price → Input: brand, year,
mileage; Output: price.
• Other: Steering angle of self-driving car.
Why Supervised Learning?
• Prediction: Guess outcomes for new cases
• Knowledge extraction: Rules reveal insights
• Compression: Simple rules > storing all data
• Outlier detection: Spot fraud or unusual cases
Unsupervised Learning
• No teacher, only raw data.
• Goal: discover hidden patterns/groups.
• Example: Customers group into 'tech lovers' or
'budget shoppers'.
• Analogy: Kids form groups in a playground.
Clustering
• Groups similar data.
• Examples: Customer segmentation, document
grouping, gene expression.
• No predefined labels → groups emerge
automatically.
Reinforcement Learning
• Learning by trial & error.
• Learner interacts with environment.
• Actions → Rewards or punishments.
• Examples: Chess, robot navigation, self-driving
cars.
Other Powers of ML
• Outlier/Novelty Detection: Spot rare/new
cases (fraud).
• Knowledge extraction: Understand
differences.
• Compression: Capture essence of data in
rules.
The History of ML
• From statistics (inference, estimation).
• Pattern recognition in engineering.
• Neural networks (1980s).
• Support Vector Machines (1990s).
• Deep Learning (2010s).
Related Topics
• High-Performance Computing: GPUs,
distributed training.
• Data Privacy & Security: anonymization,
fairness.
• Interpretability & Trust: Explainable AI.
• Data Science: ML + computing + ethics.
Key Takeaways
• ML = teaching computers to learn from data.
• Use ML when rules are unknown but data
exists.
• Types: Supervised, Unsupervised,
Reinforcement.
• Applications are everywhere.
Q&A
• Questions?
