Nicolò Dal Fabbro, Arman Adibi, Aritra Mitra, George J. Pappas
{"title":"异步多代理 TD 学习的有限时间分析","authors":"Nicolò Dal Fabbro, Arman Adibi, Aritra Mitra, George J. Pappas","doi":"arxiv-2407.20441","DOIUrl":null,"url":null,"abstract":"Recent research endeavours have theoretically shown the beneficial effect of\ncooperation in multi-agent reinforcement learning (MARL). In a setting\ninvolving $N$ agents, this beneficial effect usually comes in the form of an\n$N$-fold linear convergence speedup, i.e., a reduction - proportional to $N$ -\nin the number of iterations required to reach a certain convergence precision.\nIn this paper, we show for the first time that this speedup property also holds\nfor a MARL framework subject to asynchronous delays in the local agents'\nupdates. In particular, we consider a policy evaluation problem in which\nmultiple agents cooperate to evaluate a common policy by communicating with a\ncentral aggregator. In this setting, we study the finite-time convergence of\n\\texttt{AsyncMATD}, an asynchronous multi-agent temporal difference (TD)\nlearning algorithm in which agents' local TD update directions are subject to\nasynchronous bounded delays. Our main contribution is providing a finite-time\nanalysis of \\texttt{AsyncMATD}, for which we establish a linear convergence\nspeedup while highlighting the effect of time-varying asynchronous delays on\nthe resulting convergence rate.","PeriodicalId":501315,"journal":{"name":"arXiv - CS - Multiagent Systems","volume":"51 1","pages":""},"PeriodicalIF":0.0000,"publicationDate":"2024-07-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Finite-Time Analysis of Asynchronous Multi-Agent TD Learning\",\"authors\":\"Nicolò Dal Fabbro, Arman Adibi, Aritra Mitra, George J. Pappas\",\"doi\":\"arxiv-2407.20441\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Recent research endeavours have theoretically shown the beneficial effect of\\ncooperation in multi-agent reinforcement learning (MARL). In a setting\\ninvolving $N$ agents, this beneficial effect usually comes in the form of an\\n$N$-fold linear convergence speedup, i.e., a reduction - proportional to $N$ -\\nin the number of iterations required to reach a certain convergence precision.\\nIn this paper, we show for the first time that this speedup property also holds\\nfor a MARL framework subject to asynchronous delays in the local agents'\\nupdates. In particular, we consider a policy evaluation problem in which\\nmultiple agents cooperate to evaluate a common policy by communicating with a\\ncentral aggregator. In this setting, we study the finite-time convergence of\\n\\\\texttt{AsyncMATD}, an asynchronous multi-agent temporal difference (TD)\\nlearning algorithm in which agents' local TD update directions are subject to\\nasynchronous bounded delays. Our main contribution is providing a finite-time\\nanalysis of \\\\texttt{AsyncMATD}, for which we establish a linear convergence\\nspeedup while highlighting the effect of time-varying asynchronous delays on\\nthe resulting convergence rate.\",\"PeriodicalId\":501315,\"journal\":{\"name\":\"arXiv - CS - Multiagent Systems\",\"volume\":\"51 1\",\"pages\":\"\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2024-07-29\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"arXiv - CS - Multiagent Systems\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/arxiv-2407.20441\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"arXiv - CS - Multiagent Systems","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/arxiv-2407.20441","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
Finite-Time Analysis of Asynchronous Multi-Agent TD Learning
Recent research endeavours have theoretically shown the beneficial effect of
cooperation in multi-agent reinforcement learning (MARL). In a setting
involving $N$ agents, this beneficial effect usually comes in the form of an
$N$-fold linear convergence speedup, i.e., a reduction - proportional to $N$ -
in the number of iterations required to reach a certain convergence precision.
In this paper, we show for the first time that this speedup property also holds
for a MARL framework subject to asynchronous delays in the local agents'
updates. In particular, we consider a policy evaluation problem in which
multiple agents cooperate to evaluate a common policy by communicating with a
central aggregator. In this setting, we study the finite-time convergence of
\texttt{AsyncMATD}, an asynchronous multi-agent temporal difference (TD)
learning algorithm in which agents' local TD update directions are subject to
asynchronous bounded delays. Our main contribution is providing a finite-time
analysis of \texttt{AsyncMATD}, for which we establish a linear convergence
speedup while highlighting the effect of time-varying asynchronous delays on
the resulting convergence rate.