Stanford CS229 Machine Learning | Spring 2026 | Lecture 1: Introduction

This course provides a technical foundation in machine learning, covering mathematical principles and three primary paradigms—supervised, unsupervised, and reinforcement learning—while addressing the practical necessities of computational efficiency and ethical development.

• Course prerequisites — Enrollment requires proficiency in probability and linear algebra, as the curriculum emphasizes mathematical derivations and algorithm foundations over software implementation [02:16].

• AI tool policy — Students may use AI tools as collaborators, but utilizing them to generate direct assignment answers is prohibited to ensure students process the material themselves [02:58].

• Core definition — Learning occurs when a computer program improves its performance on a task through exposure to experience or data, evaluated by a success metric [06:08].

• Supervised learning — This method maps inputs to outputs using labeled data, covering tasks like numerical regression and category classification [08:44].

• Unsupervised learning — Algorithms analyze raw data lacking labels to discover inherent patterns, groupings, or structural relationships [20:02].

• Reinforcement learning — Systems perform sequential actions where current choices affect future outcomes, used in robotics and optimizing language model outputs via reward feedback [28:30].

• System efficiency — Optimizing hardware and software interaction is essential, as minor performance gains provide large time and cost advantages [35:54].

• What distinguishes regression from classification in supervised learning?

• Why is reinforcement learning used for modern language models?