UMass AI&Sec SP'25 Seminar: Jonas Geiping, Increasing Trust through New Benchmarks and Measuring Model Oversight
Content

Speaker
Abstract
In this talk I want to talk about recent work in posing relevant benchmarks to modern model behavior that complement existing evaluations. I want to then talk about generalizations into broader domains, which require model oversight. We'll discuss how model oversight is limited by model similarity, and how to correctly quantify model similarity.
Bio
Jonas Geiping is a Research Group Leader at the ELLIS Institute Tübingen and the Max Planck Institute for Intelligent Systems, where he leads the Safety- & Efficiency-Aligned Learning group. His research focuses on the intersection of machine learning safety and efficiency, addressing challenges such as data poisoning, adversarial attacks, watermarking, and privacy in federated learning. Jonas has published extensively in top-tier venues including NeurIPS, ICML, and ICLR. He earned his Ph.D. from the University of Siegen, and was a postdoctoral researcher at the University of Maryland.
Host