An Empirical Investigation of Value-Based Multi-objective Reinforcement Learning for Stochastic Environments

One common approach to solve multi-objective reinforcement learning (MORL) problems is to extend conventional Q-learning by using vector Q-values in combination with a utility function. However issues can arise with this approach in the context of stochastic environments, particularly when optimisin...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Ding, Kewen, Vamplew, Peter, Foale, Cameron, Dazeley, Richard
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 Ding, Kewen
Vamplew, Peter
Foale, Cameron
Dazeley, Richard
description One common approach to solve multi-objective reinforcement learning (MORL) problems is to extend conventional Q-learning by using vector Q-values in combination with a utility function. However issues can arise with this approach in the context of stochastic environments, particularly when optimising for the Scalarised Expected Reward (SER) criterion. This paper extends prior research, providing a detailed examination of the factors influencing the frequency with which value-based MORL Q-learning algorithms learn the SER-optimal policy for an environment with stochastic state transitions. We empirically examine several variations of the core multi-objective Q-learning algorithm as well as reward engineering approaches, and demonstrate the limitations of these methods. In particular, we highlight the critical impact of the noisy Q-value estimates issue on the stability and convergence of these algorithms.
doi_str_mv 10.48550/arxiv.2401.03163
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2401_03163</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2401_03163</sourcerecordid><originalsourceid>FETCH-LOGICAL-a673-5fe99e5dd4553ed9eddf1070775230b2266daccb02ca47873cc9262cb671ff3f3</originalsourceid><addsrcrecordid>eNotj9FKwzAYRnPjhUwfwCvzAq1p0iTr5RxVBx2CDm_L3-TPjLTpSLOib-82vfrg8HHgEHJXsLxcSskeIH77OeclK3ImCiWuSb8KtB4OPnoDPd2EGafk95D8GOjo6Af0R8weYUJLt8c--WzsvtAkPyN9Qx_cGA0OGBJtEGLwYU9PiL6n0XzCyWRoHWYfx3D-TDfkykE_4e3_Lsjuqd6tX7Lm9XmzXjUZKC0y6bCqUFpbSinQVmitK5hmWksuWMe5UhaM6Rg3UOqlFsZUXHHTKV04J5xYkPs_7SW3PUQ_QPxpz9ntJVv8As8YVOQ</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>An Empirical Investigation of Value-Based Multi-objective Reinforcement Learning for Stochastic Environments</title><source>arXiv.org</source><creator>Ding, Kewen ; Vamplew, Peter ; Foale, Cameron ; Dazeley, Richard</creator><creatorcontrib>Ding, Kewen ; Vamplew, Peter ; Foale, Cameron ; Dazeley, Richard</creatorcontrib><description>One common approach to solve multi-objective reinforcement learning (MORL) problems is to extend conventional Q-learning by using vector Q-values in combination with a utility function. However issues can arise with this approach in the context of stochastic environments, particularly when optimising for the Scalarised Expected Reward (SER) criterion. This paper extends prior research, providing a detailed examination of the factors influencing the frequency with which value-based MORL Q-learning algorithms learn the SER-optimal policy for an environment with stochastic state transitions. We empirically examine several variations of the core multi-objective Q-learning algorithm as well as reward engineering approaches, and demonstrate the limitations of these methods. In particular, we highlight the critical impact of the noisy Q-value estimates issue on the stability and convergence of these algorithms.</description><identifier>DOI: 10.48550/arxiv.2401.03163</identifier><language>eng</language><subject>Computer Science - Learning</subject><creationdate>2024-01</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>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,780,885</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2401.03163$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2401.03163$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Ding, Kewen</creatorcontrib><creatorcontrib>Vamplew, Peter</creatorcontrib><creatorcontrib>Foale, Cameron</creatorcontrib><creatorcontrib>Dazeley, Richard</creatorcontrib><title>An Empirical Investigation of Value-Based Multi-objective Reinforcement Learning for Stochastic Environments</title><description>One common approach to solve multi-objective reinforcement learning (MORL) problems is to extend conventional Q-learning by using vector Q-values in combination with a utility function. However issues can arise with this approach in the context of stochastic environments, particularly when optimising for the Scalarised Expected Reward (SER) criterion. This paper extends prior research, providing a detailed examination of the factors influencing the frequency with which value-based MORL Q-learning algorithms learn the SER-optimal policy for an environment with stochastic state transitions. We empirically examine several variations of the core multi-objective Q-learning algorithm as well as reward engineering approaches, and demonstrate the limitations of these methods. In particular, we highlight the critical impact of the noisy Q-value estimates issue on the stability and convergence of these algorithms.</description><subject>Computer Science - Learning</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2024</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotj9FKwzAYRnPjhUwfwCvzAq1p0iTr5RxVBx2CDm_L3-TPjLTpSLOib-82vfrg8HHgEHJXsLxcSskeIH77OeclK3ImCiWuSb8KtB4OPnoDPd2EGafk95D8GOjo6Af0R8weYUJLt8c--WzsvtAkPyN9Qx_cGA0OGBJtEGLwYU9PiL6n0XzCyWRoHWYfx3D-TDfkykE_4e3_Lsjuqd6tX7Lm9XmzXjUZKC0y6bCqUFpbSinQVmitK5hmWksuWMe5UhaM6Rg3UOqlFsZUXHHTKV04J5xYkPs_7SW3PUQ_QPxpz9ntJVv8As8YVOQ</recordid><startdate>20240106</startdate><enddate>20240106</enddate><creator>Ding, Kewen</creator><creator>Vamplew, Peter</creator><creator>Foale, Cameron</creator><creator>Dazeley, Richard</creator><scope>AKY</scope><scope>GOX</scope></search><sort><creationdate>20240106</creationdate><title>An Empirical Investigation of Value-Based Multi-objective Reinforcement Learning for Stochastic Environments</title><author>Ding, Kewen ; Vamplew, Peter ; Foale, Cameron ; Dazeley, Richard</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a673-5fe99e5dd4553ed9eddf1070775230b2266daccb02ca47873cc9262cb671ff3f3</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2024</creationdate><topic>Computer Science - Learning</topic><toplevel>online_resources</toplevel><creatorcontrib>Ding, Kewen</creatorcontrib><creatorcontrib>Vamplew, Peter</creatorcontrib><creatorcontrib>Foale, Cameron</creatorcontrib><creatorcontrib>Dazeley, Richard</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Ding, Kewen</au><au>Vamplew, Peter</au><au>Foale, Cameron</au><au>Dazeley, Richard</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>An Empirical Investigation of Value-Based Multi-objective Reinforcement Learning for Stochastic Environments</atitle><date>2024-01-06</date><risdate>2024</risdate><abstract>One common approach to solve multi-objective reinforcement learning (MORL) problems is to extend conventional Q-learning by using vector Q-values in combination with a utility function. However issues can arise with this approach in the context of stochastic environments, particularly when optimising for the Scalarised Expected Reward (SER) criterion. This paper extends prior research, providing a detailed examination of the factors influencing the frequency with which value-based MORL Q-learning algorithms learn the SER-optimal policy for an environment with stochastic state transitions. We empirically examine several variations of the core multi-objective Q-learning algorithm as well as reward engineering approaches, and demonstrate the limitations of these methods. In particular, we highlight the critical impact of the noisy Q-value estimates issue on the stability and convergence of these algorithms.</abstract><doi>10.48550/arxiv.2401.03163</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2401.03163
ispartof
issn
language eng
recordid cdi_arxiv_primary_2401_03163
source arXiv.org
subjects Computer Science - Learning
title An Empirical Investigation of Value-Based Multi-objective Reinforcement Learning for Stochastic Environments
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-23T22%3A39%3A49IST&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=An%20Empirical%20Investigation%20of%20Value-Based%20Multi-objective%20Reinforcement%20Learning%20for%20Stochastic%20Environments&rft.au=Ding,%20Kewen&rft.date=2024-01-06&rft_id=info:doi/10.48550/arxiv.2401.03163&rft_dat=%3Carxiv_GOX%3E2401_03163%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