Solving Continual Offline RL through Selective Weights Activation on Aligned Spaces

Continual offline reinforcement learning (CORL) has shown impressive ability in diffusion-based lifelong learning systems by modeling the joint distributions of trajectories. However, most research only focuses on limited continual task settings where the tasks have the same observation and action s...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Hu, Jifeng, Huang, Sili, Shen, Li, Yang, Zhejian, Hu, Shengchao, Tang, Shisong, Chen, Hechang, Chang, Yi, Tao, Dacheng, Sun, Lichao
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 Hu, Jifeng
Huang, Sili
Shen, Li
Yang, Zhejian
Hu, Shengchao
Tang, Shisong
Chen, Hechang
Chang, Yi
Tao, Dacheng
Sun, Lichao
description Continual offline reinforcement learning (CORL) has shown impressive ability in diffusion-based lifelong learning systems by modeling the joint distributions of trajectories. However, most research only focuses on limited continual task settings where the tasks have the same observation and action space, which deviates from the realistic demands of training agents in various environments. In view of this, we propose Vector-Quantized Continual Diffuser, named VQ-CD, to break the barrier of different spaces between various tasks. Specifically, our method contains two complementary sections, where the quantization spaces alignment provides a unified basis for the selective weights activation. In the quantized spaces alignment, we leverage vector quantization to align the different state and action spaces of various tasks, facilitating continual training in the same space. Then, we propose to leverage a unified diffusion model attached by the inverse dynamic model to master all tasks by selectively activating different weights according to the task-related sparse masks. Finally, we conduct extensive experiments on 15 continual learning (CL) tasks, including conventional CL task settings (identical state and action spaces) and general CL task settings (various state and action spaces). Compared with 16 baselines, our method reaches the SOTA performance.
doi_str_mv 10.48550/arxiv.2410.15698
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2410_15698</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2410_15698</sourcerecordid><originalsourceid>FETCH-arxiv_primary_2410_156983</originalsourceid><addsrcrecordid>eNpjYJA0NNAzsTA1NdBPLKrILNMzMgEKGJqaWVpwMgQH5-eUZealKzjn55Vk5pUm5ij4p6XlZOalKgT5KJRkFOWXpmcoBKfmpCaXZJalKoSnZqZnlBQrOIK4iSWZ-XkKQOSYk5mel5qiEFyQmJxazMPAmpaYU5zKC6W5GeTdXEOcPXTB1scXFGXmJhZVxoOcEQ92hjFhFQDDLj2l</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>Solving Continual Offline RL through Selective Weights Activation on Aligned Spaces</title><source>arXiv.org</source><creator>Hu, Jifeng ; Huang, Sili ; Shen, Li ; Yang, Zhejian ; Hu, Shengchao ; Tang, Shisong ; Chen, Hechang ; Chang, Yi ; Tao, Dacheng ; Sun, Lichao</creator><creatorcontrib>Hu, Jifeng ; Huang, Sili ; Shen, Li ; Yang, Zhejian ; Hu, Shengchao ; Tang, Shisong ; Chen, Hechang ; Chang, Yi ; Tao, Dacheng ; Sun, Lichao</creatorcontrib><description>Continual offline reinforcement learning (CORL) has shown impressive ability in diffusion-based lifelong learning systems by modeling the joint distributions of trajectories. However, most research only focuses on limited continual task settings where the tasks have the same observation and action space, which deviates from the realistic demands of training agents in various environments. In view of this, we propose Vector-Quantized Continual Diffuser, named VQ-CD, to break the barrier of different spaces between various tasks. Specifically, our method contains two complementary sections, where the quantization spaces alignment provides a unified basis for the selective weights activation. In the quantized spaces alignment, we leverage vector quantization to align the different state and action spaces of various tasks, facilitating continual training in the same space. Then, we propose to leverage a unified diffusion model attached by the inverse dynamic model to master all tasks by selectively activating different weights according to the task-related sparse masks. Finally, we conduct extensive experiments on 15 continual learning (CL) tasks, including conventional CL task settings (identical state and action spaces) and general CL task settings (various state and action spaces). Compared with 16 baselines, our method reaches the SOTA performance.</description><identifier>DOI: 10.48550/arxiv.2410.15698</identifier><language>eng</language><subject>Computer Science - Learning</subject><creationdate>2024-10</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>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,781,886</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2410.15698$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2410.15698$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Hu, Jifeng</creatorcontrib><creatorcontrib>Huang, Sili</creatorcontrib><creatorcontrib>Shen, Li</creatorcontrib><creatorcontrib>Yang, Zhejian</creatorcontrib><creatorcontrib>Hu, Shengchao</creatorcontrib><creatorcontrib>Tang, Shisong</creatorcontrib><creatorcontrib>Chen, Hechang</creatorcontrib><creatorcontrib>Chang, Yi</creatorcontrib><creatorcontrib>Tao, Dacheng</creatorcontrib><creatorcontrib>Sun, Lichao</creatorcontrib><title>Solving Continual Offline RL through Selective Weights Activation on Aligned Spaces</title><description>Continual offline reinforcement learning (CORL) has shown impressive ability in diffusion-based lifelong learning systems by modeling the joint distributions of trajectories. However, most research only focuses on limited continual task settings where the tasks have the same observation and action space, which deviates from the realistic demands of training agents in various environments. In view of this, we propose Vector-Quantized Continual Diffuser, named VQ-CD, to break the barrier of different spaces between various tasks. Specifically, our method contains two complementary sections, where the quantization spaces alignment provides a unified basis for the selective weights activation. In the quantized spaces alignment, we leverage vector quantization to align the different state and action spaces of various tasks, facilitating continual training in the same space. Then, we propose to leverage a unified diffusion model attached by the inverse dynamic model to master all tasks by selectively activating different weights according to the task-related sparse masks. Finally, we conduct extensive experiments on 15 continual learning (CL) tasks, including conventional CL task settings (identical state and action spaces) and general CL task settings (various state and action spaces). Compared with 16 baselines, our method reaches the SOTA performance.</description><subject>Computer Science - Learning</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2024</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNpjYJA0NNAzsTA1NdBPLKrILNMzMgEKGJqaWVpwMgQH5-eUZealKzjn55Vk5pUm5ij4p6XlZOalKgT5KJRkFOWXpmcoBKfmpCaXZJalKoSnZqZnlBQrOIK4iSWZ-XkKQOSYk5mel5qiEFyQmJxazMPAmpaYU5zKC6W5GeTdXEOcPXTB1scXFGXmJhZVxoOcEQ92hjFhFQDDLj2l</recordid><startdate>20241021</startdate><enddate>20241021</enddate><creator>Hu, Jifeng</creator><creator>Huang, Sili</creator><creator>Shen, Li</creator><creator>Yang, Zhejian</creator><creator>Hu, Shengchao</creator><creator>Tang, Shisong</creator><creator>Chen, Hechang</creator><creator>Chang, Yi</creator><creator>Tao, Dacheng</creator><creator>Sun, Lichao</creator><scope>AKY</scope><scope>GOX</scope></search><sort><creationdate>20241021</creationdate><title>Solving Continual Offline RL through Selective Weights Activation on Aligned Spaces</title><author>Hu, Jifeng ; Huang, Sili ; Shen, Li ; Yang, Zhejian ; Hu, Shengchao ; Tang, Shisong ; Chen, Hechang ; Chang, Yi ; Tao, Dacheng ; Sun, Lichao</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-arxiv_primary_2410_156983</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2024</creationdate><topic>Computer Science - Learning</topic><toplevel>online_resources</toplevel><creatorcontrib>Hu, Jifeng</creatorcontrib><creatorcontrib>Huang, Sili</creatorcontrib><creatorcontrib>Shen, Li</creatorcontrib><creatorcontrib>Yang, Zhejian</creatorcontrib><creatorcontrib>Hu, Shengchao</creatorcontrib><creatorcontrib>Tang, Shisong</creatorcontrib><creatorcontrib>Chen, Hechang</creatorcontrib><creatorcontrib>Chang, Yi</creatorcontrib><creatorcontrib>Tao, Dacheng</creatorcontrib><creatorcontrib>Sun, Lichao</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Hu, Jifeng</au><au>Huang, Sili</au><au>Shen, Li</au><au>Yang, Zhejian</au><au>Hu, Shengchao</au><au>Tang, Shisong</au><au>Chen, Hechang</au><au>Chang, Yi</au><au>Tao, Dacheng</au><au>Sun, Lichao</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Solving Continual Offline RL through Selective Weights Activation on Aligned Spaces</atitle><date>2024-10-21</date><risdate>2024</risdate><abstract>Continual offline reinforcement learning (CORL) has shown impressive ability in diffusion-based lifelong learning systems by modeling the joint distributions of trajectories. However, most research only focuses on limited continual task settings where the tasks have the same observation and action space, which deviates from the realistic demands of training agents in various environments. In view of this, we propose Vector-Quantized Continual Diffuser, named VQ-CD, to break the barrier of different spaces between various tasks. Specifically, our method contains two complementary sections, where the quantization spaces alignment provides a unified basis for the selective weights activation. In the quantized spaces alignment, we leverage vector quantization to align the different state and action spaces of various tasks, facilitating continual training in the same space. Then, we propose to leverage a unified diffusion model attached by the inverse dynamic model to master all tasks by selectively activating different weights according to the task-related sparse masks. Finally, we conduct extensive experiments on 15 continual learning (CL) tasks, including conventional CL task settings (identical state and action spaces) and general CL task settings (various state and action spaces). Compared with 16 baselines, our method reaches the SOTA performance.</abstract><doi>10.48550/arxiv.2410.15698</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2410.15698
ispartof
issn
language eng
recordid cdi_arxiv_primary_2410_15698
source arXiv.org
subjects Computer Science - Learning
title Solving Continual Offline RL through Selective Weights Activation on Aligned Spaces
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-16T20%3A57%3A31IST&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=Solving%20Continual%20Offline%20RL%20through%20Selective%20Weights%20Activation%20on%20Aligned%20Spaces&rft.au=Hu,%20Jifeng&rft.date=2024-10-21&rft_id=info:doi/10.48550/arxiv.2410.15698&rft_dat=%3Carxiv_GOX%3E2410_15698%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