報告題目： Loss surfaces of neural networks and random matrix theory.
報 告 人： Joseph Najnudel教授 University of Bristol
報告時間： 2023年6月24日 9：00-10：00
報告地點： 數學樓第二報告廳
校内聯系人：韓月才 hanyc@jlu.edu.cn
報告摘要：In this talk, based on joint work with Baskerville, Keating and Mezzadri, we study the link between universal properties of random matrices and some kind of neural networks, in particular with respect to the behavior of the critical points of their loss functions, and the corresponding Hessian matrices.
報告人簡介：Joseph Najnudel obtained his Ph.D. from the Université Paris VI in 2007. He is currently a Associate Professor at the University of Bristol, UK. His major research interests lie in random unitary matrices (in particular the Circular Unitary Ensemble), random permutation matrices and related infinite-dimensional limiting objects, link between random matrix theory and number theory, link between random matrix theory and mathematical physics，limiting measures constructed by modifying the Brownian motion or more general stochastic processes and the Brownian local times and polymer models.
