Time: 3-4 pm Eastern Time, Feb 20 (Friday), 2026

Location: MSEE 112

Zoom Link: https://purdue-edu.zoom.us/j/95468302127

Coffee and snacks will be provide.

Discovering and Controlling AI Safety Risks in Foundation Models: A Probabilistic Perspective

Abstract:

As foundation models, including large language models and multimodal models, are increasingly deployed in complex and high-stakes settings, ensuring their safety has become more important than ever. In this talk, I present a probabilistic perspective on AI safety: safety risks are treated as structured distributions to be discovered and controlled, rather than isolated failures to be patched. I first introduce probabilistic red-teaming methods that characterize distributions of failures, revealing systematic safety risks that standard evaluation often misses. I then describe probabilistic defense methods that control model behavior during deployment by adaptively steering generation toward constraint-aligned distributions. By unifying failure discovery and behavior control under a probabilistic perspective, this talk highlights a distributional approach for understanding and managing safety risks in foundation models.

Speaker:

Ruqi Zhang is an Assistant Professor in the Department of Computer Science at Purdue University. Her research focuses on probabilistic machine learning, generative modeling, and trustworthy AI. Prior to joining Purdue, she was a postdoctoral researcher at the Institute for Foundations of Machine Learning (IFML) at the University of Texas at Austin. She received her Ph.D. from Cornell University. Dr. Zhang has been a key organizer of the Symposium on Probabilistic Machine Learning. She has served as an Area Chair and Editor for ML conferences and journals, including ICML, NeurIPS, ICLR, AISTATS, UAI, and TMLR. Her contributions have been recognized with several honors, including AAAI New Faculty Highlights, Amazon Research Award, Spotlight Rising Star in Data Science, Seed for Success Acorn Award, and Ross-Lynn Research Scholar.

Organizers: Ziran Wang (ziran@purdue.edu), Yan Gu (yangu@purdue.edu), Yu She (shey@purdue.edu)