讲座时间：2024年4月4日14:00-15:00
讲座地点：腾讯会议号：683 574 363
主讲人简介：
何洪津，宁波大学数学与统计学院教授。2012年6月博士毕业于南京师范大学计算数学专业，导师孙文瑜教授和韩德仁教授。主要研究兴趣为最优化理论、算法和相关应用，相关成果发表在NumerischeMathematik、Inverse Problems、Journal of Scientific Computing等计算数学和运筹优化的权威期刊。主持和参与国家、省自然科学基金多项，2017年入选浙江省高校中青年学科带头人。
讲座内容简介：
Title:Solving saddle point problems via primal-dual algorithms: from symmetric viewpoint
Abstract:Convex-concave saddle point problems frequently arise from image processing and machine learning. In this talk, we will introduce two primal-dual algorithms (denoted by SPIDA and DEPDA) from symmetric viewpoint. It is noteworthy that the SPIDA is versatile in the sense that it enjoys an algorithmic framework covering some existing algorithms such as the classical augmentedLagrangianmethod (ALM), linearized ALM, and Jacobian splitting algorithm for linearly constrained optimization problems. Besides, SPIDA allows us to derive some customized versions so that it works as efficiently as possible for structured optimization problems. A series of numerical experiments on the matrix game, basis pursuit, robust principalcomponent analysis, and image restoration demonstrate that our SPIDA and DEPDA works well on synthetic and real-world datasets.
