Powered by Indico v3.0.3
Abstract:
The presentation first outlines the formation of jets and traditional reconstruction algorithms in high-energy physics experiments. It highlights the challenges faced by machine learning methods in tackling jet discrimination, such as the reliance on hand-crafted parameters, the breaking of Lorentz symmetry, and sensitivity to "pileup" effects. To address these challenges, the presentation explores several advanced machine learning models. An initial exploration involved treating the entire collider event as an image and applying the Mask R-CNN model for object detection and instance segmentation to identify Higgs and top quark jets. Furthermore, the report introduces the use of Graph Neural Networks, which represent particles in an event as graph nodes and classify them based on their proximity. The core of the presentation is the detailed introduction of a novel multi-task learning model based on the Mamba architecture. This model unifies instance segmentation, classification, and kinematic regression into a single system, enabling the simultaneous identification of various primary jets—such as those from Higgs bosons, top quarks, W/Z bosons, b quarks, and light-flavor quarks/gluons—along with their internal sub-jet structures.
About the speaker:
Jinmian Li received his PhD in physics at the Institute of Theoretical Physics, Chinese Academy of Sciences in 2014. He worked in the CoEPP center of Adelaide University (2014-2016) and Korea Institute for Advance Study (2016-2018) as a postdoc researcher. He joined Sichuan University in 2018 as an associate professor. His research interests mainly focus on the phenomenology of new physics beyond the Standard Model, including dark matter physics, collider physics and GUT model building.
zoom link: https://zoom.us/j/93013209092?pwd=aPyqTh2N2i80FVat1ag4Jxe712txjJ.1
zoom ID: 9301 3209 092
password: 251112