CS229: Machine Learning
by Stanford University
Stanford's graduate ML course, notes and lectures public — now running all the way to transformers.
Overview
CS229 is Stanford's graduate-level introduction to machine learning, and the version taught on campus in 2026 is considerably broader than the classic recordings most people have seen. The Summer 2026 offering is taught by Jehangir Amjad and Anand Avati and meets twice a week for 1 hour 45 minutes in the NVIDIA Auditorium; the Spring 2026 lecture recordings posted publicly to YouTube include Neural Networks (architecture), Representation Learning, Transformers and In-Context Learning, and Basic Concepts in RL with Policy Gradients, so the syllabus now runs from classical statistical learning through to the machinery behind modern language models. The consolidated course notes are a free public PDF covering supervised learning (linear and locally weighted regression, Newton's method, logistic regression, the perceptron, generalized linear models, Gaussian discriminant analysis, naive Bayes, SVMs and kernels), learning theory, regularization and model selection, debugging and error analysis, decision trees and ensembles, and neural networks; unsupervised learning (k-means, expectation maximization, factor analysis, PCA and ICA); and reinforcement learning including adaptive control, LQR, DDP and LQG. Stanford states the prerequisites explicitly: Python and NumPy at CS106A/CS106B level, probability theory, and multivariable calculus and linear algebra to the standard of MATH51 or CS205L. Lecture notes and recordings are public, but problem sets and the Ed and Canvas course spaces are restricted to enrolled students, so self-learners get the theory and must source their own exercises.
At a Glance
- Topic
- ML
- Level
- Advanced
- Format
- Course
- Cost
- Free
- Duration
- Full university course — 16+ recorded lectures of ~1h45m plus the public notes; self-paced
- Provider
- Stanford University
- Hands-on
- No
- Certificate
- None
What You’ll Learn
- ✓Derive linear, logistic and locally weighted regression from maximum-likelihood principles
- ✓Work through generalized linear models, GDA, naive Bayes and the exponential family
- ✓Understand SVMs, kernels and the margin argument behind them mathematically
- ✓Apply learning theory, bias-variance and model selection to diagnose real models
- ✓Implement unsupervised methods: k-means, EM, factor analysis, PCA and ICA
- ✓Follow the 2026 lectures on neural network architecture, representation learning and transformers
- ✓Reason about reinforcement learning, policy gradients, LQR and adaptive control
Highlights
- •The consolidated main-notes PDF is free, public and covers the whole syllabus in one document
- •Spring 2026 lecture recordings are on YouTube, so the modern transformer material is available too
- •Mathematical derivations rather than framework tutorials — the theory transfers to any library
- •Taught to the same standard as the on-campus course, at no cost and with no enrolment
- •Honest caveat: problem sets, Ed and Canvas are restricted to enrolled Stanford students
Who It’s For
Best For
- ✓Engineers who want to understand why machine-learning methods work, not just call them
- ✓Anyone preparing to read machine-learning papers without hitting a mathematics wall
- ✓Self-learners with a strong calculus and linear algebra background
- ✓Practitioners moving from applied deep learning back to statistical foundations
Prerequisites
- •Multivariable calculus and linear algebra to MATH51 or CS205L standard
- •Probability theory, including distributions and expectations
- •Python and NumPy proficiency at CS106A/CS106B level
- •Willingness to work through derivations rather than code tutorials
FAQ
What is CS229: Machine Learning?
Stanford's canonical graduate machine-learning course, with its lecture notes and recent lecture recordings free to the public. It derives the algorithms mathematically rather than teaching library calls, and the 2026 syllabus now reaches transformers, representation learning and policy-gradient reinforcement learning. Work through it and you can read a machine-learning paper and reconstruct the method from first principles.
Is CS229: Machine Learning free?
CS229: Machine Learning is free to access.
What level is CS229: Machine Learning for?
CS229: Machine Learning is aimed at a advanced audience. Recommended background: Multivariable calculus and linear algebra to MATH51 or CS205L standard, Probability theory, including distributions and expectations, Python and NumPy proficiency at CS106A/CS106B level, Willingness to work through derivations rather than code tutorials.
How long does CS229: Machine Learning take?
Expect roughly Full university course — 16+ recorded lectures of ~1h45m plus the public notes; self-paced. Most learners work through it at their own pace.
What will I learn from CS229: Machine Learning?
You'll learn: Derive linear, logistic and locally weighted regression from maximum-likelihood principles; Work through generalized linear models, GDA, naive Bayes and the exponential family; Understand SVMs, kernels and the margin argument behind them mathematically; Apply learning theory, bias-variance and model selection to diagnose real models; Implement unsupervised methods: k-means, EM, factor analysis, PCA and ICA; Follow the 2026 lectures on neural network architecture, representation learning and transformers; Reason about reinforcement learning, policy gradients, LQR and adaptive control.
Topics
Sources
This page was written from 4 sources, 2 on domains other than cs229.stanford.edu.