arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2311.02426math.OCcs.LG
收藏

带长期非凸约束的在线非凸优化

Online Non-convex Optimization with Long-term Non-convex Constraints

Department of Electrical and Computer Engineering(电气与计算机工程系);Johns Hopkins University(约翰霍普金斯大学);International Business School(国际商务学院);Suzhou Xi’an Jiaotong-Liverpool University(苏州市西安交通大学利物浦大学);Department of Data and Systems Engineering(数据与系统工程系);HKU Musketeers Foundation(HKU Musketeers基金会);Institute of Data Science(数据科学研究所);The University of Hong Kong(香港大学)

Shijie Pan, Jianyu Xu, Wenjie Huang

AI总结:

针对目标和约束均非凸的在线长期约束优化问题,提出结合拉格朗日重构、原始方向指数随机扰动与对偶方向强凹对数正则化的Follow-the-Perturbed-Leader类算法,首次实现次线性累积悔度,在河流污染源识别任务上性能优于现有方法。

摘要:

本文提出并分析了一种新型的Follow-the-Perturbed-Leader(跟随扰动领导者)类算法,用于以在线方式求解通用的长期约束优化问题,其中目标函数与约束函数由无感知对抗式生成,且不一定是凸的。该算法基于拉格朗日重构,创新性地在原始方向和对偶方向整合了随机扰动与正则化:1)原始方向采用指数分布的随机扰动以处理非凸性;2)对偶空间采用强凹对数正则化以处理约束违反情况。基于提出的期望静态累积悔度,在温和的利普希茨连续性假设下,该算法展现出在线可学习性,首次实现了此类问题的次线性累积悔度复杂度。该算法被应用于求解长期(极值)约束的河流污染源识别问题,验证了理论结果,且相比现有方法表现出更优性能。

英文摘要:

A novel Follow-the-Perturbed-Leader type algorithm is proposed and analyzed for solving general long-term constrained optimization problems in an online manner, where the target and constraint functions are oblivious adversarially generated and not necessarily convex. The algorithm is based on Lagrangian reformulation and innovatively integrates random perturbations and regularizations in primal and dual directions: 1). exponentially distributed random perturbations in the primal direction to handle non-convexity, and 2). strongly concave logarithmic regularizations in the dual space to handle constraint violations. Based on a proposed expected static cumulative regret, and under mild Lipschitz continuity assumption, the algorithm demonstrates the online learnability, achieving the first sublinear cumulative regret complexity for this class of problems. The proposed algorithm is applied to tackle a long-term (extreme value) constrained river pollutant source identification problem, validate the theoretical results and exhibit superior performance compared to existing methods.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑