Mathematical Aspects of Deep Learning – Intro

This spring I will be teaching a course on mathematical aspects of deep learning.
This blog will contain summary of the lectures by me and students taking the course.

The official course information at MIT is:

18.177 Topics in Stochastic Processes Mossel, Elchanan M 9-12 2-147

I plan to cover essentially 3 different topics:

A. The expressive power of deep networks.

B. Computational aspects of deep learning.

C. Simple probabilistic models of deep learning.

Two disclaimers:

1. The theoretical understanding of deep learning is limited. There is definitely no mathematical theory that explains why deep learning works well, but some questions related to deep learning can be formulated and analyzed mathematically. Those aspects will be the topic of this class.

2. I have done no work in topics A. and B. above. Part of my goal in the class is to understand these topics better.

Later this week I will add some references to the papers I hope to cover.