From Noise to Image

Every image model you have ever used is the descendant of a handful of mathematical ideas — an autoencoder that learned to imagine, an adversarial game between two networks, and a process that destroys pictures with noise and then learns to run the destruction backwards. This book is the chronological story of those ideas: what each one actually says (in math, and in plain English), what it enabled, and why it moved markets.

It is written as a reference: every chapter is dated, every claim cited to the original paper, every equation explained twice. Read it front to back as a history, or jump by concept — the tags follow each idea across all four books.