From Approximation to Emergence — unified deep learning theory
New arXiv book manuscript bridges classical learning theory (approximation, optimization, generalization) to modern phenomena (overparameterization, transformers, scaling laws, emergence).
Organizes scattered results into a single narrative: each theory is framed by what it controls, its assumptions, and what it leaves unexplained.
Targets researchers and graduate students seeking a coherent map of deep learning foundations.