主题:凸复合优化问题一种带自适应步长的线性临近点算法及其应用 Linearized proximal algorithms with adaptive stepsizes for convex composite optimization with applications
主讲人:浙江大学李冲教授
主持人:经济555000jc赌船孟开文
时间:2021年4月21日(周三)10:00
地点:通博楼B412
主办单位:经济555000jc赌船 科研处
主讲人简介:
李冲,现任浙江大学数学系教授,博士生导师。1995年获浙江大学博士学位,现任Pure and Applied Functional Analysis杂志、Linear and Nonlinear Analysis杂志、《高等学校计算数学学报》杂志编委。主要从事Banach空间理论、非光滑分析、非线性逼近与优化、数值泛函分析等领域的研究。先后主持中国国家自然科学基金、西班牙及南非国家自然科学基金等二十余项,出版专著1部,在SCI期刊上发表论文100余篇, 特别是在优化理论的顶级刊物SIAM J Optim.,SIAM J. Control Optim.和Math. Program等权威期刊上发表论文逾20篇。 曾获浙江省教委科技进步奖一、二等奖等奖励,先后被评为享受国务院政府特殊津贴专家、原商业部有突出贡献的中青年专家、江苏省青蓝工程优秀骨干教师。江苏省第七届青年科学家、教育部优秀骨干教师等,2004年获教育部首届新世纪优秀人才计划资助。
内容提要:
报告人将首先给出了复合凸优化问题的定义及其应用领域,接着从高斯-牛顿法求解的困难点引出线性临近点算法(LPA),并给出了LPA局部超线性收敛的条件。进一步对全局LPA及非精确LPA做了分析。最后,通过对传感器网络定位问题求解,将LPA与SDR (SemiDefiniteRelaxation)方法做了比较与分析,给出数值算例表明LPA算法具有适用范围广、收敛速度快等优点。
In this talk, we will propose an inexact linearized proximal algorithm with an adaptive stepsize, together with its globalized version based on the back-tracking line-search, to solve the convex composite optimization problem. Under the assumptions of local weak sharp minima of orderp(p≥1) for the outer convex function and a quasi-regularity condition for the inclusion problem associated to the inner function, we establish the superlinear/quadratic convergence results for the proposed algorithms. Compared to the linearized proximal algorithms with a constant stepsize proposed in [1], our algorithms own broader applications and higher convergence rates. Numerical applications to the nonnegative inverse eigenvalue problem and the wireless sensor network localization problem indicate that the proposed algorithms are more efficient and robust, and outperform the algorithms proposed in [1] and some popular algorithms for relevant problems.