A Graduate ML Course with an Obsessive Changelog

What it does

This is a public archive of lecture slides, homework PDFs, and mathematical notes for DS-GA 1003, a graduate machine learning and computational statistics course. The README functions primarily as a year-by-year changelog, documenting which topics were expanded—like the representer theorem and elastic net—and which were dropped for lack of time, such as the geometric derivation of SVMs and most of the AdaBoost lecture.

The interesting bit

Most course repos dump a syllabus and call it a day; this one treats pedagogy like software maintenance. You can watch the professor replace a one-hour Lagrangian duality crash course with a ten-minute summary, or justify dropping a “crowd-pleaser” geometric SVM derivation because it simply wasn’t worth the lecture time anymore.

Key highlights

• Extensive lecture notes and homework on regularization paths, SVM duality, and kernel methods
• Detailed mathematical notes translating classic papers (e.g., Zou and Hastie’s elastic net) into the course’s notation
• Year-over-year diffs showing the addition of backpropagation homework, Bayesian bandit notes, and EM algorithm diagrams
• Materials hosted as PDFs via GitHub Pages, with the repo acting as a structured index and changelog

Caveats

• This is a content archive, not a runnable codebase; expect PDFs and slide decks, not notebooks or libraries
• The README is comprehensive but can feel like reading release notes for a course you didn’t attend

Verdict

Worth bookmarking if you teach graduate ML or want a rigorous, mathematically heavy reference on topics like the representer theorem and convex optimization. Skip it if you’re looking for interactive notebooks or a MOOC-style guided experience.