Debuggable Deep Networks

Eric Wong (MIT)

Eric Wong is an incoming Assistant Professor at the University of Pennsylvania. Currently, he is a postdoctoral researcher in CSAIL at MIT, advised by Aleksander Madry. His work focuses on the foundations of reliable data-driven systems, building on elements of machine learning and optimization to diagnose, understand, and develop systems for real-world settings. He is a 2020 Siebel Scholar and received an honorable mention for his thesis on the robustness of deep networks to adversarial examples at CMU.

Short Abstract: How do you debug a network? What does it even mean to be debuggable? The fundamental design of current deep learning models is in opposition to the traditional debugging workflows for engineering. In this talk, I will discuss how we can re-design models to be inherently debuggable, and properly diagnose their decision processes and failure modes.