Dynamic regret for decentralized online bandit gradient descent with local steps

IF 3.7 3区计算机科学 Q2 AUTOMATION & CONTROL SYSTEMS

Journal of The Franklin Institute-engineering and Applied Mathematics Pub Date : 2025-02-01 DOI:10.1016/j.jfranklin.2025.107530

Honglei Liu, Baoyong Zhang, Deming Yuan

{"title":"Dynamic regret for decentralized online bandit gradient descent with local steps","authors":"Honglei Liu, Baoyong Zhang, Deming Yuan","doi":"10.1016/j.jfranklin.2025.107530","DOIUrl":null,"url":null,"abstract":"<div><div>In this paper, we focus on a decentralized online convex optimization problem over a multi-agent system, where each agent is equipped with a time-varying objective function. To handle the communication bottleneck and reduce the communication costs, we consider the method of local steps, where the agents communicate with their neighbors after performing local gradient descent steps. Under bandit feedback, we develop the Local-Decentralized Online Bandit Gradient Descent (Local-DOBGD) algorithm, which combines local steps and gradient descent. The performance of the developed algorithm is analyzed and the dynamic regret bound <span><math><mrow><mi>O</mi><mfenced><mrow><msqrt><mrow><mi>T</mi><mrow><mo>(</mo><mn>1</mn><mo>+</mo><msub><mrow><mi>P</mi></mrow><mrow><mi>T</mi></mrow></msub><mo>)</mo></mrow></mrow></msqrt></mrow></mfenced></mrow></math></span> is obtained, which is concerned with time horizon <span><math><mi>T</mi></math></span> and path-length <span><math><msub><mrow><mi>P</mi></mrow><mrow><mi>T</mi></mrow></msub></math></span>. Finally, we provide a numerical example to verify the effectiveness of the Local-DOBGD algorithm.</div></div>","PeriodicalId":17283,"journal":{"name":"Journal of The Franklin Institute-engineering and Applied Mathematics","volume":"362 4","pages":"Article 107530"},"PeriodicalIF":3.7000,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of The Franklin Institute-engineering and Applied Mathematics","FirstCategoryId":"94","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0016003225000249","RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"AUTOMATION & CONTROL SYSTEMS","Score":null,"Total":0}

引用次数: 0

Abstract

In this paper, we focus on a decentralized online convex optimization problem over a multi-agent system, where each agent is equipped with a time-varying objective function. To handle the communication bottleneck and reduce the communication costs, we consider the method of local steps, where the agents communicate with their neighbors after performing local gradient descent steps. Under bandit feedback, we develop the Local-Decentralized Online Bandit Gradient Descent (Local-DOBGD) algorithm, which combines local steps and gradient descent. The performance of the developed algorithm is analyzed and the dynamic regret bound

O (\sqrt{T (1 + P_{T})})

is obtained, which is concerned with time horizon

T

and path-length

P_{T}

. Finally, we provide a numerical example to verify the effectiveness of the Local-DOBGD algorithm.

查看原文本刊更多论文

求助全文

约1分钟内获得全文求助全文

来源期刊

Journal of The Franklin Institute-engineering and Applied Mathematics 工程技术-工程：电子与电气

CiteScore

7.30

自引率

14.60%

发文量

586

审稿时长

6.9 months

期刊介绍： The Journal of The Franklin Institute has an established reputation for publishing high-quality papers in the field of engineering and applied mathematics. Its current focus is on control systems, complex networks and dynamic systems, signal processing and communications and their applications. All submitted papers are peer-reviewed. The Journal will publish original research papers and research review papers of substance. Papers and special focus issues are judged upon possible lasting value, which has been and continues to be the strength of the Journal of The Franklin Institute.