Abstract

First-order proximal splittingmethods are widely used in many fields through science and engineering, such ascompressive sensing, signal/image processing, data science, machine learningand statistics, to name a few. The goal of this talk is to establish the localconvergence analysis of first-order methods when the involved functions arepartly smooth relative to an active manifold. We show that all these methodscorrectly identify the active manifolds in finite time, and then enter a locallinear convergence regime, which is characterize precisely based on thegeometry of the underlying smooth manifold. The obtained result is verified byseveral concrete numerical experiments arising from compressed sensing,signal/image processing and machine learning.

Time

12月24日（周一）14:00-15:00

Speaker

Jingwei Liang iscurrently a postdoc research associate at Department of Applied Mathematics andTheoretical Physics, University of Cambridge. He obtained his PhD in appliedmathematics from Normandy University France. Before that, he received hismaster degree in applied mathematics in 2013 from Shanghai Jiao TongUniversity, and bachelor degree in Electronic and Information Engineering fromNanjing University of Posts and Telecommunications. His current researchinterests mainly focus on non-smooth optimization and signal/image processing.

Venue

信息管理与工程学院308室

上海财经大学（第三教学楼西侧）

上海市杨浦区武东路100号