02·Three learning paradigms: supervised, unsupervised, reinforcement

Introduction

The three fundamental learning modes differ in what the model "experiences" during training. Supervised: pairs (X, y) — labeled examples (e.g., ImageNet, MNIST). Unsupervised: only X — the model seeks structure (clustering, dimensionality reduction, density modeling). Reinforcement: an agent in an environment receiving rewards for actions (AlphaGo, ChatGPT-RLHF, robotics). The lesson clarifies the definitions, illustrates them with reference algorithms (k-means, KNN, DQN, PPO), and explains hybrids: self-supervised, semi-supervised, contrastive.