Machine learning models today attain impressive accuracy on many benchmark tasks. Yet, these models also remain remarkably brittle—small perturbations of natural inputs can completely degrade their performance. 

Why is this the case?

In this talk, we take a closer look at this brittleness, and examine how it can, in part, be attributed to the fact that our models often make decisions very differently from humans. Viewing neural networks as feature extractors, we study how features extracted by neural networks may diverge from those used by humans, and how adversarially robust models seem to make progress towards bridging this gap.