北华大学学报（自然科学版）2024，Vol.25Issue(1)：8-16,9.DOI:10.11713/j.issn.1009-4822.2024.01.002
基于BB步长的近端随机递归动量算法
A Proximal Stochstic Recursive Momentum Algorithm Using Barzilai-Borwein Stepsize
摘要
Abstract
An algorithm for solving nonconvex and nonsmooth composite optimization problems is researched.Firstly,a stochastic variance reduction algorithm with BB step size(ProxSTORM-BB)for nonconvex and nonsmooth composite optimization problems is proposed by combining the proximal stochastic recursive momentum algorithm and improved BB step size.In the iterative process,the algorithm dynamically adjusts the step size to improve the computational efficiency.And it is insensitive to the selection of the initial step size,which solves the problem that the parameter tuning is difficult.Then,the convergence of the algorithm is proved under suitable assumptions.Finally,the effectiveness of the proposed algorithm is verified by numerical experiments.关键词
BB步长/近端随机递归动量算法/非凸非光滑复合优化问题Key words
Barzilai-Borwein stepsize/proximal stochastic recursive momentum algorithm/nonconvex and nonsmooth composite optimization problems分类
数理科学引用本文复制引用
钱玉香,赵勇,杨帆..基于BB步长的近端随机递归动量算法[J].北华大学学报（自然科学版）,2024,25(1):8-16,9.基金项目
重庆市自然科学基金项目(CSTB2022NSCQ). （CSTB2022NSCQ）
