News
July 2023: The recording of the tutorial is available below.
June 2023: We have released the slides used in the tutorial. Link.
Overview
The past decade has seen significant advancements in applied deep learning for tasks such as vision, NLP, as well as in deep learning theory. However, these two areas of research have evolved mostly in isolation from each other, resulting in missed connections and ideas. This tutorial aims to bridge the gap between the empirical performance of neural networks and deep learning theory. In other words, we aim to make recent deep learning theory developments accessible to vision researchers and encourage them to design new architectures and algorithms for practical tasks. The goal is to help computer vision researchers to better understand deep learning theory and apply it to design new theoretically-principled networks that can lead to breakthroughs.
Schedule Detail
Tentative schedule
-
9.00 AM
Introduction and preliminaries
-
9.15 AM
Robustness in DL -- Descent Directions
-
10.15 AM
Break
-
10.30 AM
Mathematical foundations in DL theory
-
11.30 AM
Applications to computer vision
