太阳成集团tyc7111cc学术报告

Landscape analysis of nonconvex models in phase retrieval

黄猛 博士

（香港科技大学）

报告时间： 2020年12月17日 (星期四) 15：00-16：00

腾讯会议 ID：388 971 960 

报告摘要：One efficient way to solve phase retrieval problem is through nonlinear least squares, which is a nonconvex optimization with numerous local minimas. However, simple gradient descent algorithms often work remarkably well and not get trapped in a local minimum in practice. To explain the success of these algorithms, we analyze the geometric landscape structure of the loss functions and show that for several nonconvex models in phase retrieval: a local minimum is also global minimum and at any saddle point there is a negative directional curvature. This structure allows a number of optimization algorithms to efficiently find a global minimizer without special initialization. This is a joint work with Prof. Jianfeng Cai, Dong Li and Yang Wang.

报告人简介：黄猛，2019年于中国科学院数学与系统科学研究院获博士学位，导师许志强研究员。2019年至今于香港科技大学从事博士后研究，导师汪扬教授。研究方向为相位恢复、低秩矩阵恢复和压缩感知。现已在包括《Applied and Computational Harmonic Analysis》,《IEEE Transactions on Information Theory》等顶级期刊发表论文数篇。

邀请人： 谢家新