An Efficient Lorentz equivariant graph neural network for jet tagging

by Dr Qi Meng (Microsoft Research Asia)

Wednesday 15 Dec 2021, 10:30 → 11:30 Asia/Shanghai

B105 (CHEP)

Machine learning methods especially deep learning have been becoming popular on jet representation in particle physics. Most of these methods focus on the handcrafted feature design or tuning the structure of existing black-box deep neural networks, while they ignore the Lorentz group equivariance, a fundamental space time symmetry in the law for jet generating. Recently, there is Lorentz equivariant deep model proposed for jet tagging, however, it is time-consuming due to the explicit reliance on the high-order tensors. In this paper, we propose an efficient LorentzNet, which directly scalarizes the tensors to realize Lorentz group symmetry. Experiments on two representative jet tagging benchmarks show that LorentNet can achieve the best tagging accuracy under both clean and Lorentz rotated test setting. Biography: Qi Meng （孟琪） is a senior researcher in Computational and Learning Group (COLT), in Microsoft Research Asia (MSRA). Before she joined Microsoft in July 2018, she obtained her Ph.D. in probability and mathematical statistics from Peking University, supervised by Prof. Zhi-Ming Ma （马志明）, Chinese Academy of Sciences. Her main research interests include deep learning theory, learning dynamics, optimization theory and statistics. https://www.microsoft.com/en-us/research/people/meq/ Tecent Meeting： https://meeting.tencent.com/dm/m9mzbqx1WcMa Meeting ID：540-627-958