Accelerated Stochastic ADMM and Its Extensions (加速随机交替方向法及其拓展)
白建超（西北工业大学数学与统计学院）
2022年3月17日晚上20:00-22:00
腾讯会议：674 767 848
摘要：
In this talk, an Accelerated Stochastic Alternating Direction Method of Multipliers (AS-ADMM) is firstly presented for solving the separable convex optimization problem whose objective function is the sum of a possibly nonsmooth convex function and an average function of many smooth convex functions. Under proper assumptions, this AS-ADMM is shown to converge in expectation with a sublinear/linear convergence rate. We also extend the method to a symmetric AS-ADMM (SAS-ADMM) where the dual variable is updated twice with relatively larger stepsizes, and an accelerated stochastic Peaceman Rachford splitting method which enjoys indefinite proximal term and relaxation technique. Preliminary experiments indicate that SAS-ADMM performs very efficient for solving the so-called graph-guided fused lasso model in machine learning.
报告人简介：白建超，博士(后)，西北工业大学数学与统计学院副教授，硕士生导师，CSIAM大数据与人工智能专业委员会委员。主要从事机器学习和统计学习等领域的大规模优化方法、理论与应用研究，近五年以第一作者在中国数学会公布的Top杂志上发表SCI论文10余篇，主持国家自然科学基金1项。
