题目：A Framework for Analyzing Variance Reduced Stochastic Gradient Methods and a New One for Non-smooth Non-convex Optimization
时间：2023年4月18日10:00-10:45
地点：数学院425
报告人：梁经纬副教授，上海交通大学自然科学研究院
邀请人：陈亮
摘要: Over the past years, stochastic optimization methods are becoming increasingly popular in traditional areas including inverse problems and signal/image processing. In this talk, I will introduce SPRING, a novel stochastic version of proximal alternating linearized minimization (PALM) algorithm for solving a class of non-smooth and non-convex optimization problems which arise in many statistical machine learning, computer vision and imaging applications. Theoretically, I will show that our proposed method with variance-reduced stochastic gradient estimators, such as SAGA and SARAH, achieves state-of-the-art oracle complexities. Numerical experiments on sparse non-negative matrix factorization, sparse principal component analysis and blind image deconvolution are also presented to demonstrate the efficiency of our algorithm.
报告人简介： 梁经纬于 2013 年获得上海交通 大学数学硕士学位，之后于 2016 年获得法国卡昂大学数学博士学位。2017 至 2020 年， 梁经纬在英国剑桥大学理论物理与应用数学系从事博士后研究工作，并于 2020 年底加 入伦敦玛丽王后大学数学科学学院任数据科学讲师。2021 年 7 月，正式加入上海交通 大学。梁经纬的主要研究兴趣为数学图像处理，非光滑优化和数据科学等。
