报告题目：Solving inverse problems in high energy physics using deep learning
报告人：施舒哲 助理教授 （清华大学）
报告时间：2023年12月12日 上午10:00
报告地点：理8栋118学术报告厅
报告内容：
Deep Neural Networks (DNNs) have been widely employed a general and universal parameterization scheme to represent arbitrary continuous functions. The advantages of DNN lie in not only its generality but also the efficient computation of the parameter-gradients, via the Back-Propagation procedures. In ordinary deep learning problems, one is usually able to directly compare the output of DNN to the observed data, and train the DNN parameters accordingly. In some special problems, however, the relation between the observable and the underlying property of interest, represented by DNN, could be complicated or even implicit. In this talk, I will present a general discussion on how to adjust the BackPropagation algorithm according to such complicated/implicit problems. Then, I will focus on three realistic examples, the inverse Schroedinger equation solver, inverse TOV equation solver, and the inverse Laplace transformation, to discuss the application and possible limitation.。
报告人简介：
Shuzhe Shi is currently an Assistant Professor at Department of Physics, Tsinghua University. He obtained his bachelor(2012) and master(2015) degrees from Tsinghua under the supervision of Prof. Pengfei Zhuang. Then he received the Ph.D.(2018) from Indiana Univ., Bloomington, supervised by Prof. Jinfeng Liao. After that, he became PostDoc researchers at McGill Univ.(2021) and Stony Brook Univ.(2023). His research focus on novel transport phenomena in heavy ion collisions, as well as applying machine learning and quantum computation methods in nuclear physics.