报告题目：Golden Ratio Primal-Dual Hybrid GradientAlgorithm

报告人：常小凯 副教授 兰州理工大学

照片：

邀请人：刘三阳

报告时间：2021年12月3日（星期五）18:15-19:15

报告地点：腾讯会议:322 205 400

报告人简介：常小凯，博士，兰州理工大学副教授。2019年博士毕业于伟德国际BETVlCTOR，主要从事最优化方法、理论与应用研究。在Journal of Scientific Computing，Journal of Optimization Theory and Applications等期刊发表SCI论文10余篇。现主持国家自然科学基金项目和甘肃省教育厅基金项目各1项。

报告摘要：We design, analyze and test a golden ratio primal-dual hybrid gradient algorithm (GRPDHG), which is a new variant of the classical Arrow-Hurwicz method for solving bilinearsaddle point problem. The same as the popular PDHG of Chambolle and Pock, GRPDHG is fullsplitting.We propose a linesearch strategy for GRPDHG tocompute adaptive step-sizes. Measured by function value residual and constraint violation ofan equivalent constrained optimization problem, its globalergodic convergence rate, acceleration and linear convergence were obtained under some conditions.

This is a joint work with Professor Junfeng Yang (Nanjing University) and Professor Hongchao Zhang (Louisiana State University)