Instructor: Jatin Batra
Schedule: Thursdays, 2:00–3:30 PM
Room: A-238
Email: jatinbatra50@gmail.com
Textbook: Learning theory from first principles — Francis Bach (Chapter 9 and relevant parts of Chapters 3, 4, 5, 7, 8, 12, 14).

Course Description

This mini-course serves as a take-off point for research in theory of deep learning, with neural networks as the prototypical setup, more specifically, the focus will be on implicit bias of overparameterized models. Generalization power of modern deep learning cannot be easily explained as the number of pa- rameters is often of the same order as the number of training samples. In this part, we will initiate an understanding of implicit bias which captures additional properties enjoyed by natural training procedures for overpa- rameterized models that are not captured explicitly, and which could hold the key to generalization.

Weekly Outline