Reducing Conservativeness Oriented Offline Reinforcement Learning

In offline reinforcement learning, a policy learns to maximize cumulative rewards with a fixed collection of data. Towards conservative strategy, current methods choose to regularize the behavior policy or learn a lower bound of the value function. However, exorbitant conservation tends to impair th...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Zhang, Hongchang, Shao, Jianzhun, Jiang, Yuhang, He, Shuncheng, Ji, Xiangyang
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 Zhang, Hongchang
Shao, Jianzhun
Jiang, Yuhang
He, Shuncheng
Ji, Xiangyang
description In offline reinforcement learning, a policy learns to maximize cumulative rewards with a fixed collection of data. Towards conservative strategy, current methods choose to regularize the behavior policy or learn a lower bound of the value function. However, exorbitant conservation tends to impair the policy's generalization ability and degrade its performance, especially for the mixed datasets. In this paper, we propose the method of reducing conservativeness oriented reinforcement learning. On the one hand, the policy is trained to pay more attention to the minority samples in the static dataset to address the data imbalance problem. On the other hand, we give a tighter lower bound of value function than previous methods to discover potential optimal actions. Consequently, our proposed method is able to tackle the skewed distribution of the provided dataset and derive a value function closer to the expected value function. Experimental results demonstrate that our proposed method outperforms the state-of-the-art methods in D4RL offline reinforcement learning evaluation tasks and our own designed mixed datasets.
doi_str_mv 10.48550/arxiv.2103.00098
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2103_00098</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2103_00098</sourcerecordid><originalsourceid>FETCH-LOGICAL-a678-59d63b062bd1ee547eb471fbf7374057e127b5ab4afd121d1ffa2bd2aa3b09243</originalsourceid><addsrcrecordid>eNotj81qwkAUhWfTRbF9gK6cF0ic30yylFDbQiAg7sMd514Z0FEmNujbN7VdHTic78DH2JsUpamtFSvItziVSgpdCiGa-pmttxi-9zEdeHtOI-YJrnHChOPI-xwxXTHwnugYE_ItxkTnvMfT3PMOIacZfGFPBMcRX_9zwXab9137WXT9x1e77gqoXF3YJlTai0r5IBGtceiNk-TJaWeEdSiV8xa8AQpSySCJYN4qgJlqlNELtvy7fTgMlxxPkO_Dr8vwcNE_7CNFLw</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>Reducing Conservativeness Oriented Offline Reinforcement Learning</title><source>arXiv.org</source><creator>Zhang, Hongchang ; Shao, Jianzhun ; Jiang, Yuhang ; He, Shuncheng ; Ji, Xiangyang</creator><creatorcontrib>Zhang, Hongchang ; Shao, Jianzhun ; Jiang, Yuhang ; He, Shuncheng ; Ji, Xiangyang</creatorcontrib><description>In offline reinforcement learning, a policy learns to maximize cumulative rewards with a fixed collection of data. Towards conservative strategy, current methods choose to regularize the behavior policy or learn a lower bound of the value function. However, exorbitant conservation tends to impair the policy's generalization ability and degrade its performance, especially for the mixed datasets. In this paper, we propose the method of reducing conservativeness oriented reinforcement learning. On the one hand, the policy is trained to pay more attention to the minority samples in the static dataset to address the data imbalance problem. On the other hand, we give a tighter lower bound of value function than previous methods to discover potential optimal actions. Consequently, our proposed method is able to tackle the skewed distribution of the provided dataset and derive a value function closer to the expected value function. Experimental results demonstrate that our proposed method outperforms the state-of-the-art methods in D4RL offline reinforcement learning evaluation tasks and our own designed mixed datasets.</description><identifier>DOI: 10.48550/arxiv.2103.00098</identifier><language>eng</language><subject>Computer Science - Learning</subject><creationdate>2021-02</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>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,776,881</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2103.00098$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2103.00098$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Zhang, Hongchang</creatorcontrib><creatorcontrib>Shao, Jianzhun</creatorcontrib><creatorcontrib>Jiang, Yuhang</creatorcontrib><creatorcontrib>He, Shuncheng</creatorcontrib><creatorcontrib>Ji, Xiangyang</creatorcontrib><title>Reducing Conservativeness Oriented Offline Reinforcement Learning</title><description>In offline reinforcement learning, a policy learns to maximize cumulative rewards with a fixed collection of data. Towards conservative strategy, current methods choose to regularize the behavior policy or learn a lower bound of the value function. However, exorbitant conservation tends to impair the policy's generalization ability and degrade its performance, especially for the mixed datasets. In this paper, we propose the method of reducing conservativeness oriented reinforcement learning. On the one hand, the policy is trained to pay more attention to the minority samples in the static dataset to address the data imbalance problem. On the other hand, we give a tighter lower bound of value function than previous methods to discover potential optimal actions. Consequently, our proposed method is able to tackle the skewed distribution of the provided dataset and derive a value function closer to the expected value function. Experimental results demonstrate that our proposed method outperforms the state-of-the-art methods in D4RL offline reinforcement learning evaluation tasks and our own designed mixed datasets.</description><subject>Computer Science - Learning</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2021</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotj81qwkAUhWfTRbF9gK6cF0ic30yylFDbQiAg7sMd514Z0FEmNujbN7VdHTic78DH2JsUpamtFSvItziVSgpdCiGa-pmttxi-9zEdeHtOI-YJrnHChOPI-xwxXTHwnugYE_ItxkTnvMfT3PMOIacZfGFPBMcRX_9zwXab9137WXT9x1e77gqoXF3YJlTai0r5IBGtceiNk-TJaWeEdSiV8xa8AQpSySCJYN4qgJlqlNELtvy7fTgMlxxPkO_Dr8vwcNE_7CNFLw</recordid><startdate>20210226</startdate><enddate>20210226</enddate><creator>Zhang, Hongchang</creator><creator>Shao, Jianzhun</creator><creator>Jiang, Yuhang</creator><creator>He, Shuncheng</creator><creator>Ji, Xiangyang</creator><scope>AKY</scope><scope>GOX</scope></search><sort><creationdate>20210226</creationdate><title>Reducing Conservativeness Oriented Offline Reinforcement Learning</title><author>Zhang, Hongchang ; Shao, Jianzhun ; Jiang, Yuhang ; He, Shuncheng ; Ji, Xiangyang</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a678-59d63b062bd1ee547eb471fbf7374057e127b5ab4afd121d1ffa2bd2aa3b09243</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2021</creationdate><topic>Computer Science - Learning</topic><toplevel>online_resources</toplevel><creatorcontrib>Zhang, Hongchang</creatorcontrib><creatorcontrib>Shao, Jianzhun</creatorcontrib><creatorcontrib>Jiang, Yuhang</creatorcontrib><creatorcontrib>He, Shuncheng</creatorcontrib><creatorcontrib>Ji, Xiangyang</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Zhang, Hongchang</au><au>Shao, Jianzhun</au><au>Jiang, Yuhang</au><au>He, Shuncheng</au><au>Ji, Xiangyang</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Reducing Conservativeness Oriented Offline Reinforcement Learning</atitle><date>2021-02-26</date><risdate>2021</risdate><abstract>In offline reinforcement learning, a policy learns to maximize cumulative rewards with a fixed collection of data. Towards conservative strategy, current methods choose to regularize the behavior policy or learn a lower bound of the value function. However, exorbitant conservation tends to impair the policy's generalization ability and degrade its performance, especially for the mixed datasets. In this paper, we propose the method of reducing conservativeness oriented reinforcement learning. On the one hand, the policy is trained to pay more attention to the minority samples in the static dataset to address the data imbalance problem. On the other hand, we give a tighter lower bound of value function than previous methods to discover potential optimal actions. Consequently, our proposed method is able to tackle the skewed distribution of the provided dataset and derive a value function closer to the expected value function. Experimental results demonstrate that our proposed method outperforms the state-of-the-art methods in D4RL offline reinforcement learning evaluation tasks and our own designed mixed datasets.</abstract><doi>10.48550/arxiv.2103.00098</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2103.00098
ispartof
issn
language eng
recordid cdi_arxiv_primary_2103_00098
source arXiv.org
subjects Computer Science - Learning
title Reducing Conservativeness Oriented Offline Reinforcement Learning
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-22T11%3A33%3A27IST&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=Reducing%20Conservativeness%20Oriented%20Offline%20Reinforcement%20Learning&rft.au=Zhang,%20Hongchang&rft.date=2021-02-26&rft_id=info:doi/10.48550/arxiv.2103.00098&rft_dat=%3Carxiv_GOX%3E2103_00098%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