Types of Machine Learning

1. Supervised Learning

The algorithm learns from labelled training data — each example has an input and a known correct output.

Classification — predict a category:

Email spam detection: input = email text → output = SPAM / NOT_SPAM
Image recognition: input = image pixels → output = class label

Regression — predict a number:

House price prediction: input = size, location → output = price
Sales forecasting: input = historical data → output = future sales

Common algorithms: Linear Regression, Logistic Regression, Decision Trees, Random Forests, SVMs, Neural Networks

2. Unsupervised Learning

The algorithm finds patterns in unlabelled data — no correct answers provided.

Clustering — group similar items:

Customer segmentation: group customers by purchasing behaviour
Document clustering: group news articles by topic

Dimensionality reduction — compress data while keeping structure:

PCA (Principal Component Analysis): reduce 100 features to 2 for visualisation

Common algorithms: K-Means, DBSCAN, PCA, Autoencoders

3. Reinforcement Learning

An agent learns by interacting with an environment — receives rewards for good actions and penalties for bad ones.

AlphaGo: agent = neural network, environment = Go board, reward = winning
Self-driving car: agent = car, environment = road, reward = reaching destination safely

Key concepts: policy, reward function, Q-learning, PPO

4. Self-Supervised / Foundation Models

Modern LLMs are trained with a self-supervised objective: predict the next token in a sequence. No manual labels needed — the data labels itself.

Input:  "The quick brown fox jumps"
Target: "over the lazy dog"

This lets models train on internet-scale data without expensive human labelling.

Quick reference

Type	Labelled data?	Example use case
Supervised	Yes	Spam filter, fraud detection
Unsupervised	No	Customer segmentation
Reinforcement	Reward signal	Game playing, robotics
Self-supervised	Self-generated	LLMs (GPT, Claude)

Interview Tip

For product/backend engineer interviews, you're expected to know these distinctions conceptually — not to implement them. The practical question is: "What type of ML problem is this?" Being able to frame a business problem as supervised/unsupervised demonstrates ML fluency.