Deep Learning Applications in LHAASO-KM2A

24 Oct 2026, 09:00
20m
Tianfu Cosmic Ray Research Center

Tianfu Cosmic Ray Research Center

No. 1500 Kezhi Road, Tianfu New Area, Chengdu, China

Speaker

Chao Yang (IHEP)

Description

This report presents recent progress in the application of machine learning techniques to data analysis with the Kilometer Square Array (KM2A) of the Large High Altitude Air Shower Observatory (LHAASO), focusing on two key tasks: gamma–proton discrimination and air-shower direction reconstruction. We have developed deep learning models for these tasks and conducted extensive validation using real observational data, including the Crab Nebula as a standard candle and several representative gamma-ray sources from the first LHAASO source catalog. These studies aim to evaluate the effectiveness and potential of machine learning approaches in practical physics analyses.

Furthermore, we have investigated semi-supervised learning strategies to better exploit the large amount of unlabeled observational data. Our results indicate that pretraining on large-scale unlabeled datasets can accelerate convergence during subsequent training on downstream tasks, offering a promising approach to improving the utilization of experimental data. In parallel, we are conducting studies on model interpretability to understand the underlying physical information and mechanisms learned by deep learning models for event discrimination and direction reconstruction. These efforts aim to establish a clearer connection between model predictions and the physical characteristics of extensive air showers.

This report will summarize our recent findings and discuss the future prospects of machine learning in high-energy gamma-ray astronomy and data analysis.

Authors

Chao Yang (IHEP) Prof. 曼奇 阮 (IHEP) 曼麟 常 (高能所) 永峰 朱

Presentation materials

There are no materials yet.