Bo Li (UIUC) - Trustworthy Federated Learning: Robustness, Fairness, Privacy, and Their Interconnections

Part of the DSI Autumn 2022 Distinguished Speaker Series.

Advances in machine learning have led to rapid and widespread deployment of ML algorithms for safety-critical applications, such as autonomous driving and medical diagnostics. Current machine learning systems, however, assume that training and test data follow the same, or similar, distributions, and do not consider active adversaries manipulating either distribution. Recent work has demonstrated that motivated adversaries can circumvent ML detection models at test time through evasion attacks, or inject well-crafted malicious instances into training data to induce errors during inference through poisoning attacks, especially in the distributed setting. In this talk, I will describe my recent research about security, privacy, and fairness problems in federated learning, with a focus on certifiably robust federated learning against training-time attacks, fairness, and the interconnection between robustness and privacy in federated learning. I will also discuss other defense principles towards developing practical trustworthy federated learning systems with guarantees.

Join us for lunch at 12; Dr. Li’s talk will begin at 12:30. This talk will also be broadcast via Zoom. Please register to receive viewing information.