🖨️ Printing Instructions: Press Ctrl/Cmd + P and select "Save as PDF".

Machine Learning Basics

Classification of AI, Supervised & Unsupervised Learning, and Beyond

Part 1: Classification of AI

Classification of AI

Artificial Narrow Intelligence (ANI): AI designed for a specific task (e.g., image recognition, playing chess). This is all of AI today.
Artificial General Intelligence (AGI): A hypothetical AI that can perform any intellectual task a human can. Does not exist yet.
Artificial Super Intelligence (ASI): A hypothetical AI that surpasses human intelligence in all areas. A topic of much debate.

Approaches of AI (Non-Inclusive)

Symbolic / Logic-Based: Uses rules and symbol manipulation (e.g., Expert Systems).
Statistical / Probabilistic: Uses probability theory to make decisions (e.g., Bayesian Networks, Statistical Learning).
Computational (Connectionist): Inspired by the brain, uses neural networks (e.g., Deep Learning).
Integrative: Combines multiple approaches.

AI Research Topics

Machine Learning
Computer Vision
Natural Language Processing (LLMs)
Reinforcement Learning
Knowledge Representation and Reasoning
Planning and Scheduling
Search Methodologies

The AI / ML / DL Hierarchy

Artificial Intelligence (AI): The broadest field, encompassing all techniques that enable machines to mimic human intelligence.
Machine Learning (ML): A subset of AI. Systems that learn from data without being explicitly programmed.
Deep Learning (DL): A subset of ML. Uses multi-layered neural networks to learn complex patterns.

Part 2: Introduction to Machine Learning

What is "Learning" for a Machine?

"Machine Learning is the science of getting computers to act without being explicitly programmed." — Arthur Samuel, 1959.
Instead of writing rules, we provide data and let the machine discover the rules.

The 4 Ingredients of Machine Learning

1. Data: The raw material. The experience the machine learns from. (e.g., images, emails, stock prices).
2. Model: The "brain" or the algorithm. A mathematical representation of the world. (e.g., a neural network, a decision tree).
3. Objective Function (Loss): The "teacher." A function that tells the model how well it's doing. The goal is to minimize this.
4. Learning Algorithm: The method used to "train" the model to achieve the objective (e.g., Gradient Descent).

Categorizing ML Techniques

We categorize ML techniques based on the kind of Data and Objective they use.
Main paradigms: Supervised, Unsupervised, Reinforcement Learning.

Part 3: Supervised Learning

What is Supervised Learning?

Definition: The task of learning a function $f(X) \rightarrow y$ that maps an input $X$ to an output $y$ based on example input-output pairs.
Keywords: Labeled data, training data, ground truth.

The Supervised Learning Loop

1. Input $X$ is fed into the Model.
2. Model produces an Output (prediction).
3. Loss Function compares output to the Targeted Output (ground truth).
4. Learning Algorithm adjusts the model based on the loss.
5. Repeat until the loss is minimized.

Two Major Tasks of Supervised Learning

Classification: Predicting a discrete category. Question: "What class does this belong to?" (e.g., spam vs. not spam, cat vs. dog).
Regression: Predicting a continuous value. Question: "How much?" or "How many?" (e.g., predicting house prices, stock prices).

Technique 1: Linear Regression

Goal: Model the relationship between a dependent variable ($y$) and one or more independent variables ($X$) by fitting a straight line.
The model learns the optimal slope and intercept to minimize the prediction error.

Interactive Demo: linear_regression_demo.html

Technique 2: k-Nearest Neighbor (kNN)

A simple, intuitive classification algorithm.
Step 1: Find the $k$ nearest data points (neighbors) to the query point.
Step 2: Classify the query point by a majority vote of its neighbors.

Interactive Demo: knn_demo.html

Part 4: Unsupervised Learning

What is Unsupervised Learning?

Definition: You only have input data ($X$) and no corresponding output labels. The goal is to find hidden patterns or intrinsic structures.
Keywords: Unlabeled data, hidden structure, patterns.

Classic Unsupervised Task: Clustering

Question it answers: "What are the natural groups in my data?"
Examples: Customer Segmentation, Image Segmentation, Genomics (grouping genes with similar expression patterns).

Technique 3: K-Means Clustering

A simple and popular clustering algorithm.
Step 1: Randomly initialize $K$ cluster centroids.
Step 2: Assign each data point to the nearest centroid.
Step 3: Update each centroid to be the mean of its assigned points.
Step 4: Repeat steps 2-3 until convergence (centroids stop moving).

Interactive Demo: kmeans_demo.html

Part 5: Learning with Less Supervision

The Labeled Data Bottleneck

Labeled data is the key ingredient for supervised learning, but it's expensive and time-consuming to create.
The world is full of unlabeled data (all of Wikipedia, all of YouTube).
How can we leverage this?

Semi-Supervised Learning

Concept: Use a small amount of labeled data + a large amount of unlabeled data.
The model learns the overall structure from the unlabeled data to improve its decisions on the labeled data.
Analogy: A student has an answer key for Chapter 1, but not for Chapters 2-20. They can still learn a lot from the text.

Transfer Learning

Concept: Take a model pre-trained on a massive dataset and adapt it for a more specific task.
Analogy: A master chef doesn't need to re-learn how to hold a knife to learn a new recipe. They transfer their skills.
Supervised Fine-Tuning (SFT): The process of adapting a pre-trained model using your smaller, labeled dataset. Extremely common today.

Self-Supervised Learning (The Key to LLMs)

Concept: Create a supervised learning task from the unlabeled data itself.
Analogy: Take a sentence from Wikipedia, hide a word, and ask the model to predict the missing word. The data is the sentence, the label is the hidden word!
Impact: This is the core idea behind models like GPT and Gemini. They learn the structure of language by predicting the next word.

Part 6: Reinforcement Learning

What is Reinforcement Learning?

It's not about labeled data; it's about learning to make a sequence of decisions.
Analogy: Training a dog. You don't show it a flashcard. You say "sit," and when it sits, you give it a treat (reward). Over time, it learns a policy to maximize its treats.

Key Components of RL

Agent: The learner or decision-maker (the dog, a robot, a game-playing AI).
Environment: The world the agent interacts with (a room, a maze, a game of Chess).
Action: A move the agent can make.
Reward: The feedback from the environment (positive or negative).
Policy: The agent's strategy for choosing actions.

RL Examples

Robotics: Learning to walk or grasp objects.
Game Playing: AlphaGo defeating the world's best Go player.
Resource Management: Optimizing cooling systems in data centers (Google DeepMind).

Summary

Key Takeaways

Supervised Learning: Labeled Data (Classification, Regression).
Unsupervised Learning: Unlabeled Data (Clustering).
Reinforcement Learning: No data labels, just an Environment (Agent, Rewards, Policy).
Semi/Self/Transfer Learning: Powerful bridges that reduce the need for expensive labeled data.