Why long model-based rollouts are no reason for bad Q-value estimates

This paper explores the use of model-based offline reinforcement learning with long model rollouts. While some literature criticizes this approach due to compounding errors, many practitioners have found success in real-world applications. The paper aims to demonstrate that long rollouts do not nece...

Ausführliche Beschreibung

Gespeichert in:

Bibliographische Detailangaben
Hauptverfasser:	Wissmann, Philipp, Hein, Daniel, Udluft, Steffen, Tresp, Volker
Format:	Artikel
Sprache:	eng
Schlagworte:	Computer Science - Learning
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	Wissmann, Philipp Hein, Daniel Udluft, Steffen Tresp, Volker
description	This paper explores the use of model-based offline reinforcement learning with long model rollouts. While some literature criticizes this approach due to compounding errors, many practitioners have found success in real-world applications. The paper aims to demonstrate that long rollouts do not necessarily result in exponentially growing errors and can actually produce better Q-value estimates than model-free methods. These findings can potentially enhance reinforcement learning techniques.
doi_str_mv	10.48550/arxiv.2407.11751
format	Article
fullrecord	<record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2407_11751</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2407_11751</sourcerecordid><originalsourceid>FETCH-arxiv_primary_2407_117513</originalsourceid><addsrcrecordid>eNqFzbEKwjAQgOEsDqI-gJP3Ao2NNtRdKq6C4Biu9KqFtCeXtNi3V4u707_88Cm1NqnODtamW5RXM-hdlubamNyauSpujxE8d3douSKflBioAmHvuY8BUAg6BiEM3EHNAiVWcEkG9D0Bhdi0GCks1axGH2j160JtTsX1eE4m0D3ls8novrCb4P3_4w3_NDgr</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>Why long model-based rollouts are no reason for bad Q-value estimates</title><source>arXiv.org</source><creator>Wissmann, Philipp ; Hein, Daniel ; Udluft, Steffen ; Tresp, Volker</creator><creatorcontrib>Wissmann, Philipp ; Hein, Daniel ; Udluft, Steffen ; Tresp, Volker</creatorcontrib><description>This paper explores the use of model-based offline reinforcement learning with long model rollouts. While some literature criticizes this approach due to compounding errors, many practitioners have found success in real-world applications. The paper aims to demonstrate that long rollouts do not necessarily result in exponentially growing errors and can actually produce better Q-value estimates than model-free methods. These findings can potentially enhance reinforcement learning techniques.</description><identifier>DOI: 10.48550/arxiv.2407.11751</identifier><language>eng</language><subject>Computer Science - Learning</subject><creationdate>2024-07</creationdate><rights>http://arxiv.org/licenses/nonexclusive-distrib/1.0</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/2407.11751$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2407.11751$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Wissmann, Philipp</creatorcontrib><creatorcontrib>Hein, Daniel</creatorcontrib><creatorcontrib>Udluft, Steffen</creatorcontrib><creatorcontrib>Tresp, Volker</creatorcontrib><title>Why long model-based rollouts are no reason for bad Q-value estimates</title><description>This paper explores the use of model-based offline reinforcement learning with long model rollouts. While some literature criticizes this approach due to compounding errors, many practitioners have found success in real-world applications. The paper aims to demonstrate that long rollouts do not necessarily result in exponentially growing errors and can actually produce better Q-value estimates than model-free methods. These findings can potentially enhance reinforcement learning techniques.</description><subject>Computer Science - Learning</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2024</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNqFzbEKwjAQgOEsDqI-gJP3Ao2NNtRdKq6C4Biu9KqFtCeXtNi3V4u707_88Cm1NqnODtamW5RXM-hdlubamNyauSpujxE8d3douSKflBioAmHvuY8BUAg6BiEM3EHNAiVWcEkG9D0Bhdi0GCks1axGH2j160JtTsX1eE4m0D3ls8novrCb4P3_4w3_NDgr</recordid><startdate>20240716</startdate><enddate>20240716</enddate><creator>Wissmann, Philipp</creator><creator>Hein, Daniel</creator><creator>Udluft, Steffen</creator><creator>Tresp, Volker</creator><scope>AKY</scope><scope>GOX</scope></search><sort><creationdate>20240716</creationdate><title>Why long model-based rollouts are no reason for bad Q-value estimates</title><author>Wissmann, Philipp ; Hein, Daniel ; Udluft, Steffen ; Tresp, Volker</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-arxiv_primary_2407_117513</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2024</creationdate><topic>Computer Science - Learning</topic><toplevel>online_resources</toplevel><creatorcontrib>Wissmann, Philipp</creatorcontrib><creatorcontrib>Hein, Daniel</creatorcontrib><creatorcontrib>Udluft, Steffen</creatorcontrib><creatorcontrib>Tresp, Volker</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Wissmann, Philipp</au><au>Hein, Daniel</au><au>Udluft, Steffen</au><au>Tresp, Volker</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Why long model-based rollouts are no reason for bad Q-value estimates</atitle><date>2024-07-16</date><risdate>2024</risdate><abstract>This paper explores the use of model-based offline reinforcement learning with long model rollouts. While some literature criticizes this approach due to compounding errors, many practitioners have found success in real-world applications. The paper aims to demonstrate that long rollouts do not necessarily result in exponentially growing errors and can actually produce better Q-value estimates than model-free methods. These findings can potentially enhance reinforcement learning techniques.</abstract><doi>10.48550/arxiv.2407.11751</doi><oa>free_for_read</oa></addata></record>
fulltext	fulltext_linktorsrc
identifier	DOI: 10.48550/arxiv.2407.11751
ispartof
issn
language	eng
recordid	cdi_arxiv_primary_2407_11751
source	arXiv.org
subjects	Computer Science - Learning
title	Why long model-based rollouts are no reason for bad Q-value estimates
url	https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-12T13%3A12%3A30IST&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=Why%20long%20model-based%20rollouts%20are%20no%20reason%20for%20bad%20Q-value%20estimates&rft.au=Wissmann,%20Philipp&rft.date=2024-07-16&rft_id=info:doi/10.48550/arxiv.2407.11751&rft_dat=%3Carxiv_GOX%3E2407_11751%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