CS229: Machine Learning

This is the theory-first original that Ng's gentler Coursera courses were built from. You work through the actual derivations and proofs, which is exactly what people heading into research or wanting to know why a method works are after. The trade-off is that it is genuinely demanding and assumes real comfort with probability and linear algebra.

The free version gives you lectures and notes but no graded assignments, no feedback and no certificate, so you supply the discipline. The recordings are from 2018 and pre-date the current wave of transformer and LLM work, though the fundamentals it teaches still underpin all of it. Pair it with a hands-on course if you also want to build.

CS229 is Andrew Ng's graduate-level Stanford ML course — the academic original that preceded his Coursera courses and goes significantly deeper into the mathematical foundations. Where the Coursera ML Specialization builds intuition and implements in Python, CS229 works through full probabilistic derivations and proofs: maximum likelihood estimation for supervised learning, EM algorithm for unsupervised learning, support vector machines and kernels, principal component analysis, reinforcement learning with MDPs and Q-learning, and the theoretical foundations of learning guarantees.