应兰州大学数学与统计学院李朋副教授邀请, 西安交通大学管理学院曾锦山教授, 将

于5月22-24号访问兰州大学并且做学术报告。

报告题目:UnlockingConvergence of ADMMinNonconvex Nonsmooth Optimization: Theory and Applications

地点：榆中校区明远楼A栋数学与统计学院五楼A528

报告时间：5月23号(星期六)上午9:00-10:00

报告摘要：The Alternating Direction Method of Multipliers (ADMM) has been widely adopted in modern machine learningsuch as sparse statistical learning, deep learning, and distributed learning,due to its remarkable computational efficiency and flexible algorithmic structure. However,many kinds oftheseproblems exhibit inherent nonconvexity and nonsmoothness, posing fundamental challenges to the convergence analysis of ADMM.In this talk, we systematically introduce a unified framework for establishing the global convergence of ADMM in nonconvex and nonsmooth optimization. We propose a set of conditions that guarantee the convergence of ADMM and show that these conditions aretight in the sense thatthe algorithm may fail to convergewhen violated.Basedon this theoreticalresult, we extend the analysis to several representative settings, including sparse statistical learning, deep learning, and federated learning, and establish global convergence guarantees for the corresponding ADMM variants.

报告人简介

曾锦山,西安交通大学管理学院青拔A类教授,博士生导师.主持国家自然科学基金面上项目、江西省重大人才计划“双千计划”、江西省杰青等项目6项,在IEEE TPAMI/TIP/ TSP、JMLR和ICML等相关领域高水平期刊或会议上发表论文90余篇,其中ESI高被引论文3篇,一篇论文入选“中国数学领域热点论文榜单前十”,单篇论文最高引用1680余次,授权发明专利20余项,荣获国际基础科学大会“前沿科学奖”、世界华人数学家联盟最佳论文奖和江西省自然科学奖二等奖等多项重要学术奖励,受邀在世界华人数学家大会上作60分钟大会报告. 指导学生获得“挑战杯”国家特等奖等国家级学科竞赛奖励30余项.主要研究方向是人工智能中优化算法理论及其应用。

甘肃省计算数学基础学科研究中心

数学与统计学院

萃英学院

2026年5月20日