Asymptotic Convergence and Performance of Multi-Agent Q-Learning Dynamics

Achieving convergence of multiple learning agents in general $N$-player games is imperative for the development of safe and reliable machine learning (ML) algorithms and their application to autonomous systems. Yet it is known that, outside the bounds of simple two-player games, convergence cannot b...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Hussain, Aamal Abbas, Belardinelli, Francesco, Piliouras, Georgios
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page
container_issue
container_start_page
container_title
container_volume
creator Hussain, Aamal Abbas
Belardinelli, Francesco
Piliouras, Georgios
description Achieving convergence of multiple learning agents in general $N$-player games is imperative for the development of safe and reliable machine learning (ML) algorithms and their application to autonomous systems. Yet it is known that, outside the bounds of simple two-player games, convergence cannot be taken for granted. To make progress in resolving this problem, we study the dynamics of smooth Q-Learning, a popular reinforcement learning algorithm which quantifies the tendency for learning agents to explore their state space or exploit their payoffs. We show a sufficient condition on the rate of exploration such that the Q-Learning dynamics is guaranteed to converge to a unique equilibrium in any game. We connect this result to games for which Q-Learning is known to converge with arbitrary exploration rates, including weighted Potential games and weighted zero sum polymatrix games. Finally, we examine the performance of the Q-Learning dynamic as measured by the Time Averaged Social Welfare, and comparing this with the Social Welfare achieved by the equilibrium. We provide a sufficient condition whereby the Q-Learning dynamic will outperform the equilibrium even if the dynamics do not converge.
doi_str_mv 10.48550/arxiv.2301.09619
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2301_09619</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2301_09619</sourcerecordid><originalsourceid>FETCH-LOGICAL-a679-a2322ef4c429c193846f1c3b7314d738ebe3c214a8bbf49ffa6b4d07fa7d81513</originalsourceid><addsrcrecordid>eNotz91KwzAcBfDceCHTB_DKvEBqvtokl6U6HVTcYPflnzQZgTUdaTfs2-umV4fDgQM_hJ4YLaQuS_oC-TteCi4oK6ipmLlHm3pahtM8ztHhZkwXnw8-OY8h9XjrcxjzANc-Bvx5Ps6R1L_7jHek9ZBTTAf8uiQYopse0F2A4-Qf_3OF9uu3ffNB2q_3TVO3BCplCHDBuQ_SSW4cM0LLKjAnrBJM9kpob71wnEnQ1gZpQoDKyp6qAKrXrGRihZ7_bm-W7pTjAHnprqbuZhI_49lHAw</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>Asymptotic Convergence and Performance of Multi-Agent Q-Learning Dynamics</title><source>arXiv.org</source><creator>Hussain, Aamal Abbas ; Belardinelli, Francesco ; Piliouras, Georgios</creator><creatorcontrib>Hussain, Aamal Abbas ; Belardinelli, Francesco ; Piliouras, Georgios</creatorcontrib><description>Achieving convergence of multiple learning agents in general $N$-player games is imperative for the development of safe and reliable machine learning (ML) algorithms and their application to autonomous systems. Yet it is known that, outside the bounds of simple two-player games, convergence cannot be taken for granted. To make progress in resolving this problem, we study the dynamics of smooth Q-Learning, a popular reinforcement learning algorithm which quantifies the tendency for learning agents to explore their state space or exploit their payoffs. We show a sufficient condition on the rate of exploration such that the Q-Learning dynamics is guaranteed to converge to a unique equilibrium in any game. We connect this result to games for which Q-Learning is known to converge with arbitrary exploration rates, including weighted Potential games and weighted zero sum polymatrix games. Finally, we examine the performance of the Q-Learning dynamic as measured by the Time Averaged Social Welfare, and comparing this with the Social Welfare achieved by the equilibrium. We provide a sufficient condition whereby the Q-Learning dynamic will outperform the equilibrium even if the dynamics do not converge.</description><identifier>DOI: 10.48550/arxiv.2301.09619</identifier><language>eng</language><subject>Computer Science - Artificial Intelligence ; Computer Science - Computer Science and Game Theory ; Computer Science - Multiagent Systems ; Mathematics - Dynamical Systems</subject><creationdate>2023-01</creationdate><rights>http://creativecommons.org/licenses/by/4.0</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,776,881</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2301.09619$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2301.09619$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Hussain, Aamal Abbas</creatorcontrib><creatorcontrib>Belardinelli, Francesco</creatorcontrib><creatorcontrib>Piliouras, Georgios</creatorcontrib><title>Asymptotic Convergence and Performance of Multi-Agent Q-Learning Dynamics</title><description>Achieving convergence of multiple learning agents in general $N$-player games is imperative for the development of safe and reliable machine learning (ML) algorithms and their application to autonomous systems. Yet it is known that, outside the bounds of simple two-player games, convergence cannot be taken for granted. To make progress in resolving this problem, we study the dynamics of smooth Q-Learning, a popular reinforcement learning algorithm which quantifies the tendency for learning agents to explore their state space or exploit their payoffs. We show a sufficient condition on the rate of exploration such that the Q-Learning dynamics is guaranteed to converge to a unique equilibrium in any game. We connect this result to games for which Q-Learning is known to converge with arbitrary exploration rates, including weighted Potential games and weighted zero sum polymatrix games. Finally, we examine the performance of the Q-Learning dynamic as measured by the Time Averaged Social Welfare, and comparing this with the Social Welfare achieved by the equilibrium. We provide a sufficient condition whereby the Q-Learning dynamic will outperform the equilibrium even if the dynamics do not converge.</description><subject>Computer Science - Artificial Intelligence</subject><subject>Computer Science - Computer Science and Game Theory</subject><subject>Computer Science - Multiagent Systems</subject><subject>Mathematics - Dynamical Systems</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2023</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotz91KwzAcBfDceCHTB_DKvEBqvtokl6U6HVTcYPflnzQZgTUdaTfs2-umV4fDgQM_hJ4YLaQuS_oC-TteCi4oK6ipmLlHm3pahtM8ztHhZkwXnw8-OY8h9XjrcxjzANc-Bvx5Ps6R1L_7jHek9ZBTTAf8uiQYopse0F2A4-Qf_3OF9uu3ffNB2q_3TVO3BCplCHDBuQ_SSW4cM0LLKjAnrBJM9kpob71wnEnQ1gZpQoDKyp6qAKrXrGRihZ7_bm-W7pTjAHnprqbuZhI_49lHAw</recordid><startdate>20230123</startdate><enddate>20230123</enddate><creator>Hussain, Aamal Abbas</creator><creator>Belardinelli, Francesco</creator><creator>Piliouras, Georgios</creator><scope>AKY</scope><scope>AKZ</scope><scope>GOX</scope></search><sort><creationdate>20230123</creationdate><title>Asymptotic Convergence and Performance of Multi-Agent Q-Learning Dynamics</title><author>Hussain, Aamal Abbas ; Belardinelli, Francesco ; Piliouras, Georgios</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a679-a2322ef4c429c193846f1c3b7314d738ebe3c214a8bbf49ffa6b4d07fa7d81513</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2023</creationdate><topic>Computer Science - Artificial Intelligence</topic><topic>Computer Science - Computer Science and Game Theory</topic><topic>Computer Science - Multiagent Systems</topic><topic>Mathematics - Dynamical Systems</topic><toplevel>online_resources</toplevel><creatorcontrib>Hussain, Aamal Abbas</creatorcontrib><creatorcontrib>Belardinelli, Francesco</creatorcontrib><creatorcontrib>Piliouras, Georgios</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv Mathematics</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Hussain, Aamal Abbas</au><au>Belardinelli, Francesco</au><au>Piliouras, Georgios</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Asymptotic Convergence and Performance of Multi-Agent Q-Learning Dynamics</atitle><date>2023-01-23</date><risdate>2023</risdate><abstract>Achieving convergence of multiple learning agents in general $N$-player games is imperative for the development of safe and reliable machine learning (ML) algorithms and their application to autonomous systems. Yet it is known that, outside the bounds of simple two-player games, convergence cannot be taken for granted. To make progress in resolving this problem, we study the dynamics of smooth Q-Learning, a popular reinforcement learning algorithm which quantifies the tendency for learning agents to explore their state space or exploit their payoffs. We show a sufficient condition on the rate of exploration such that the Q-Learning dynamics is guaranteed to converge to a unique equilibrium in any game. We connect this result to games for which Q-Learning is known to converge with arbitrary exploration rates, including weighted Potential games and weighted zero sum polymatrix games. Finally, we examine the performance of the Q-Learning dynamic as measured by the Time Averaged Social Welfare, and comparing this with the Social Welfare achieved by the equilibrium. We provide a sufficient condition whereby the Q-Learning dynamic will outperform the equilibrium even if the dynamics do not converge.</abstract><doi>10.48550/arxiv.2301.09619</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2301.09619
ispartof
issn
language eng
recordid cdi_arxiv_primary_2301_09619
source arXiv.org
subjects Computer Science - Artificial Intelligence
Computer Science - Computer Science and Game Theory
Computer Science - Multiagent Systems
Mathematics - Dynamical Systems
title Asymptotic Convergence and Performance of Multi-Agent Q-Learning Dynamics
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-28T23%3A45%3A11IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-arxiv_GOX&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=Asymptotic%20Convergence%20and%20Performance%20of%20Multi-Agent%20Q-Learning%20Dynamics&rft.au=Hussain,%20Aamal%20Abbas&rft.date=2023-01-23&rft_id=info:doi/10.48550/arxiv.2301.09619&rft_dat=%3Carxiv_GOX%3E2301_09619%3C/arxiv_GOX%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rfr_iscdi=true