Explore the critical aspects of deep learning robustness in this insightful lecture from the Deep Learning Theory Workshop and Summer School. Dive into the challenges of distribution shift and its impact on deep networks' performance. Gain new perspectives on the functioning of deep networks beyond standard generalization. Examine the transfer learning process, including optimization dynamics and improved heuristics for model transfer. Investigate the role of overparameterization and inductive biases in deep networks through the lens of robustness. Discover how studying robustness can provide valuable insights into the fundamental workings of deep learning systems.
Overview
Syllabus
Understanding the Robustness of Deep Learning
Taught by
Simons Institute