Parallelizable Approaches for Nonsmooth Optimization Problems with Orthogonality Constraints
Speaker
Prof. Xin Liu
Academy of Mathematics and Systems Science, CAS
Abstract

In previous decades, the methods for optimization problems with orthogonality constraints usually search on the manifold or the tangent space, in which explicitly or implicitly orthonormalization procedure is inevitable so that the column-wise parallelization lacks of concurrency. Recently, we propose a few efficient infeasible approaches to solve this type of problems and demonstrate their potential in parallel computing. In this talk, we extend our approaches to a few nonsmooth cases. Problems like sparse variable principle component analysis, coordinate-independent sparse estimation can be solved in parallel with high scalability.

About the Speaker

刘歆,中国科学院数学与系统科学研究院副研究员、博士生导师,国家优青获得者。2004年本科毕业于北京大学数学科学学院;2009年于中国科学院研究生院获得博士学位,导师是袁亚湘院士;毕业后留所工作至今。期间分别在德国ZIB研究所、美国RICE大学、美国纽约大学Courant研究所进行过长期访问。主要研究方向包括:正交约束矩阵优化问题,线性与非线性特征值问题,及其在电子结构计算中的应用;非线性最小二乘的算法与理论,分布式优化算法设计,及其在机器学习中的应用。刘歆在20168月获得国家自然科学基金委优秀青年科学基金;201610月获得中国运筹学会青年科技奖;20172月入选中国科学院北京分院启明星优秀人才计划。于20157月起担任《Mathematical Programming Computation》编委;201610月起担任中国运筹学会理事;20177月起担任《计算数学》编委;20186月起担任《物理学报》特约栏目编辑;20195月起担任中国工业与应用数学会副秘书长。

Date&Time
2019-11-14 4:00 PM
Location
Room: A203 Meeting Room
CSRC 新闻 CSRC News CSRC Events CSRC Seminars CSRC Divisions