The goal of this graduate-level course is to describe the mathematical aspects of the incipient theory of Deep Learning, with an emphasis on open questions and intended for graduate students aiming to focus on theoretical aspects of DL.
This year, the course is divided in two parts: (i) Theory of Supervised Shallow Neural Nets, and (ii) Beyond Supervised Shallow Learning. The first part covers the foundations of approximation, optimization and generalisation in high-dimensional supervised learning using shallow neural networks, focusing both in the linear and non-linear regimes. The second part covers miscellaneous aspects beyond, including the role of depth, namely symmetries and invariances, depth separation, robustness and unsupervised learning.
Lecture Instructor:
TA: