ECE Seminar Speaker: Krishnamurthy Dvijotham

Description

Krishnamurthy Dvijotham, a staff research scientist on the brain team at Google Research, will give a talk titled "Building Safe and Reliable AI: Perspectives from Robustness, Uncertainty Estimation, and Privacy" for the Electrical and Computer Engineering Department. Dvijotham's research focuses on building safe, reliable, and trustworthy artificial intelligence (AI), including concerns like privacy, uncertainty estimation, fairness, and formal verifiability.

Abstract:

As we make ever-accelerating progress on AI systems with powerful capabilities, the challenge of building AI systems that are safe and reliable becomes increasingly pressing. I will discuss recent work on three specific challenges: 1) How do we measure, certify, and enhance robustness of deep learning systems to test-time adversarial attacks? 2) How do we train AI systems to be aware of their uncertainty and defer back to a user when necessary? 3) How do we address these challenges while preserving the privacy of users the AI was trained on?