CPF Seminar with Shih-Chieh Hsu

Abstract: EveNet is a foundation model framework for collider event analysis that aims to unify data representation and reduce task-specific training across high-energy physics (HEP) analyses. Built on a transformer-based encoder-decoder architecture, EveNet is pre-trained on 500 million simulated proton-proton collision events using hybrid self-supervised and supervised objectives to capture intrinsic kinematic and topological structures. With minimal fine-tuning, task-specific decoders achieve state-of-the-art performance on multiple representative CMS-like analyses and maintain strong calibration when applied to real CMS Open Data. Notably, the pretrained representation generalizes effectively to out-of-distribution beyond-the-Standard-Model signals, indicating that EveNet learns fundamental physical symmetries governing event dynamics, By providing a shared universal backbone for classification, regression, and anomaly detection tasks, EveNet demonstrates how foundation models can streamline large-scale data analysis, accelerate physics interpretation, and enable scalable discovery pipelines for current and future collider experiments.

Bio: Shih-Chieh Hsu is a Professor of Physics at the University of Washington, specializing in the discovery of fundamental particles and interactions using Large Hadron Collider data and in hardware-accelerated Al for real-time scientific discovery. He directs the NSF HDR A3D3
Institute and is also an Adjunct Professor in Electrical and Computer Engineering and affiliate of the UW eScience Institute, Hsu earned his Ph,D, at UC San Diego following B,S, and M,S degrees from National Taiwan University and previously held a Chamberlain Fellowship at Lawrence Berkeley National Laboratory, His work has been recognized with a DOE Early Career Award, a UW Undergraduate Research Mentor Award, and a Chinese Translation Award for Youth Science Popularization.