报告时间：2026年9月7日（周一）10:00

报告地点：物理楼233

报  告 人：张宇雷 博士 美国华盛顿大学（西雅图）

报告摘要: 

Foundation models are changing the way machine learning is used across many fields by providing a shared pretrained model that can be adapted to a wide range of downstream tasks. In collider physics, many analyses face similar challenges in understanding complex events, reconstructing invisible particles, and extracting subtle physics information from high-dimensional data.

In this talk, I will present **EveNet**, an event-level foundation model designed to provide a common starting point for collider analyses. I will discuss how EveNet can be transferred across different physics processes and experiments, with applications ranging from event classification to generative reconstruction. Examples include $Z\to\tau\tau$ studies with archived DELPHI data at LEP, where the model improves event identification and multi-neutrino reconstruction, as well as applications in Higgs and di-Higgs analyses at the LHC.

I will also introduce our recent work combining EveNet with preference optimization, where reinforcement-learning-inspired post-training is used to improve generative inference in underconstrained neutrino reconstruction. More broadly, they illustrate how a shared foundation model can provide a common language for understanding, reconstructing, and analyzing collider events.

报告人简介：

Dr. Yulei Zhang is a postdoctoral scholar at the University of Washington and a member of the ATLAS Collaboration. He received his Ph.D. in experimental particle physics from Shanghai Jiao Tong University in 2024, jointly supervised with Université Paris Cité and the APC/CNRS in Paris.

His research focuses on Higgs physics, machine learning for particle physics, and fundamental quantum phenomena at colliders. He has played leading roles in ATLAS searches for non-resonant Higgs-boson pair production, serving as an analysis contact and machine-learning coordinator. He is currently the convener of the ATLAS Non-resonant Multileptons Subgroup and a member of the ATLAS Physics Office.

Dr. Zhang’s recent work includes EveNet, an event-level foundation model for high-energy physics, as well as machine-learning-based neutrino reconstruction for studies of quantum entanglement and Bell nonlocality in tau-lepton pairs at the LHC. He also contributes to community efforts in uncertainty-aware AI for high-energy physics and cosmology.