报告主题:Constructing splitting methods for convex optimization --- From the perspective of VI and PPA
报告人:何炳生 教授 (南方科技大学)
报告时间:2019年6月6日(周四)10:00
报告地点:校本部GJ410
邀请人:彭亚新
报告摘要:Many problems in image processing can be reduced to the linearly constrained convex optimization. The first order optimality of the linearly constrained optimization can be reformulated as the monotone variational inequalities. The convenience of studying convex optimization algorithms in the perspective of variational inequalities has been more and more recognized. From the point of view of variational inequalities, many existing methods for convex optimization, such as augmented Lagrangian Method, primal-dual hybrid gradient method, Alternating Direction Method of Multiplies, are proximal point algorithms (PPA). In last years, we have proposed a uniform framework of the prediction-correction methods whose convergence properties are closed related to PPA. In this talk, after introducing the algorithms framework, we will explain how to use this framework to construct suitable splitting methods according to the different difficulties of the sub-problems.
欢迎教师、学生参加!