What it does This is the archived syllabus and lecture notes for UC Berkeley’s Stat 212B, a graduate topics course taught by Joan Bruna in Spring 2016. It covers three chunks: convolutional networks through the lens of invariance and scattering theory, deep unsupervised learning (VAEs, GANs, maximum entropy), and a grab bag of optimization and attention models. Each lecture links to PDF slides and a curated reading list.

The interesting bit The course sits at an inflection point—GANs were brand new, normalizing flows were cutting-edge, and the syllabus treats neural networks as mathematical objects (stability, invertibility, group theory) rather than just engineering recipes. Guest lectures from Wojciech Zaremba, Soumith Chintala, and Yann Dauphin anchor it to the research frontier of that specific semester.

Key highlights

• Heavy Mallat/Bruna influence: scattering transforms and wavelet theory as foundations for understanding CNNs
• Unusually thorough reading lists, linking each lecture to 3–8 foundational papers
• Guest lectures from OpenAI and FAIR researchers on Neural Turing Machines, GANs, and attention
• Explicit coverage of non-Euclidean domains (graph CNNs) and connections to dictionary learning, LISTA, random forests
• Truncated README: the syllabus cuts off mid-sentence in the third section, suggesting the repo was never fully polished

Caveats

• Lecture PDFs are linked but not described; you can’t tell which are slides, notes, or exercises without clicking
• The repo is a syllabus, not code—expect no implementations
• Some arXiv links are to preprints that may have since been revised or superseded

Verdict Worth bookmarking if you want a mathematician’s map of deep learning circa 2016, or if you’re tracing how ideas like VAEs and GANs entered the graduate curriculum. Skip it if you need runnable code or a 2024 perspective.