Three exam-targeted study guides covering the full twelve weeks, built from the lecture notes, the instructor's handwritten note sets, the official practice papers, and every past-term quiz and end-term paper with its answer key. Each guide walks the full narrative, then drills the exact question templates that keep coming back.
The mathematics of GANs — from f-divergences and variational bounds, through Wasserstein GANs and inversion, to FID. One tightly-scoped story, told end to end.
One inequality — the ELBO — and its three descendants: EM for mixtures, the VAE with a single learned latent, and the DDPM with a thousand fixed ones.
Diffusion finished properly — score, guidance, DDIM — then the change of subject: autoregressive transformers, and the reinforcement learning that makes them behave.