数学与统计学院学术报告[2021] 043号

(高水平大学建设系列报告543号)

讲座题目: Indefinite Linearized Augmented Largrangian Method for Large-Scale Convex Optimization with Applications

讲座人：  袁景  教授（西安电子科技大学）

讲座时间：2021年5月20日下午4:00-6:00

报告地点： 汇星楼514

报告内容：Nowadays, a wide variety of real-world applications appear as convex optimization problems, esp. large-scale complicated convex programming, which motivate developments of advanced optimization and computing algorithms attaining optimum efficiently in a global sense. This, hence, stimulates an enormous spread of research topics in both mathematical analysis and optimization algorithms, such as variational analysis, convergence analysis and acceleration, high-performance parallel computing etc. We contribute this talk on our recent developments of dual optimization and linearized augmented Lagrangian methods (ALMs). We show a novel prediction-correction formulation of modern linearized ALM based on variational inequality analysis. Also, we present an analysis on the step-size of the associate proximal term, which results in an indefinite proximity regularization and speeds up convergence in numeric. Experiments on machine learning and image segmentation demonstrate superior performance of the new indefinite linearized augmented Lagrangian method and confirm our theoretical analysis on the optimal step-size and convergence rate.

报告人简历 袁景，教授，本科和硕士分别毕业于武汉大学和北京大学，博士毕业于德国海德堡大学数学与计算机科学系。于加拿大Western University计算机系完成两年博士后研究，后被Western University的Robarts研究所聘为研究员，同时被Western University的Schulich医学院聘为特聘研究教授。2016年9月回国，于西安电子科技大学数学与统计学院聘为教授，2017年获国家高层次回国人才资助。目前，发表各类学术文章超过100篇，3篇专著章节，总期刊影响因子超过200。主持国家自然科学基金面上项目一项担任美国威尼斯wns9778期刊Inverse Problems & Imaging、Elsevier期刊Smart Health编委成员。