{"id":"https://openalex.org/W3014137283","doi":"https://doi.org/10.48550/arxiv.2003.14089","title":"Leverage the Average: an Analysis of KL Regularization in RL","display_name":"Leverage the Average: an Analysis of KL Regularization in RL","publication_year":2020,"publication_date":"2020-03-31","ids":{"openalex":"https://openalex.org/W3014137283","doi":"https://doi.org/10.48550/arxiv.2003.14089","mag":"3014137283"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2003.14089","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2003.14089","pdf_url":"https://arxiv.org/pdf/2003.14089","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2003.14089","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5071735499","display_name":"Nino Vieillard","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I1291425158","display_name":"Google (United States)","ror":"https://ror.org/00njsd438","country_code":"US","type":"company","lineage":["https://openalex.org/I1291425158","https://openalex.org/I4210128969"]},{"id":"https://openalex.org/I1294671590","display_name":"Centre National de la Recherche Scientifique","ror":"https://ror.org/02feahw73","country_code":"FR","type":"government","lineage":["https://openalex.org/I1294671590"]},{"id":"https://openalex.org/I3017839442","display_name":"Institut \u00c9lie Cartan de Lorraine","ror":"https://ror.org/04rvw8791","country_code":"FR","type":"facility","lineage":["https://openalex.org/I1294671590","https://openalex.org/I1326498283","https://openalex.org/I3017839442","https://openalex.org/I4210141950","https://openalex.org/I90183372"]},{"id":"https://openalex.org/I90183372","display_name":"Universit\u00e9 de Lorraine","ror":"https://ror.org/04vfs2w97","country_code":"FR","type":"education","lineage":["https://openalex.org/I90183372"]}],"countries":["FR","US"],"is_corresponding":false,"raw_author_name":"Vieillard, Nino","raw_affiliation_strings":["Google Research , Brain Team Universit\u00e9 de Lorraine , CNRS , Inria IECL , F-54000 Nancy , France","Google,,,,,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Research , Brain Team Universit\u00e9 de Lorraine , CNRS , Inria IECL , F-54000 Nancy , France","institution_ids":["https://openalex.org/I1294671590","https://openalex.org/I3017839442","https://openalex.org/I90183372"]},{"raw_affiliation_string":"Google,,,,,","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5070075141","display_name":"Tadashi Kozuno","orcid":"https://orcid.org/0000-0002-8820-1362","observed_orcids":["https://orcid.org/0000-0002-8820-1362"]},"institutions":[{"id":"https://openalex.org/I142637625","display_name":"Okinawa Institute of Science and Technology Graduate University","ror":"https://ror.org/02qg15b79","country_code":"JP","type":"education","lineage":["https://openalex.org/I142637625"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Kozuno, Tadashi","raw_affiliation_strings":["Okinawa Institute of Science and Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Okinawa Institute of Science and Technology","institution_ids":["https://openalex.org/I142637625"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5032281855","display_name":"Bruno Scherrer","orcid":"https://orcid.org/0000-0001-6097-4489","observed_orcids":["https://orcid.org/0000-0001-6097-4489"]},"institutions":[{"id":"https://openalex.org/I1294671590","display_name":"Centre National de la Recherche Scientifique","ror":"https://ror.org/02feahw73","country_code":"FR","type":"government","lineage":["https://openalex.org/I1294671590"]},{"id":"https://openalex.org/I1326498283","display_name":"Institut national de recherche en sciences et technologies du num\u00e9rique","ror":"https://ror.org/02kvxyf05","country_code":"FR","type":"government","lineage":["https://openalex.org/I1326498283"]},{"id":"https://openalex.org/I3017839442","display_name":"Institut \u00c9lie Cartan de Lorraine","ror":"https://ror.org/04rvw8791","country_code":"FR","type":"facility","lineage":["https://openalex.org/I1294671590","https://openalex.org/I1326498283","https://openalex.org/I3017839442","https://openalex.org/I4210141950","https://openalex.org/I90183372"]},{"id":"https://openalex.org/I90183372","display_name":"Universit\u00e9 de Lorraine","ror":"https://ror.org/04vfs2w97","country_code":"FR","type":"education","lineage":["https://openalex.org/I90183372"]}],"countries":["FR"],"is_corresponding":false,"raw_author_name":"Scherrer, Bruno","raw_affiliation_strings":["Universit\u00e9 de Lorraine , CNRS , Inria IECL , F-54000 Nancy , France","French Institute for Research in Computer Science and Automation"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Universit\u00e9 de Lorraine , CNRS , Inria IECL , F-54000 Nancy , France","institution_ids":["https://openalex.org/I1294671590","https://openalex.org/I3017839442","https://openalex.org/I90183372"]},{"raw_affiliation_string":"French Institute for Research in Computer Science and Automation","institution_ids":["https://openalex.org/I1326498283"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5065100569","display_name":"Olivier Pietquin","orcid":"https://orcid.org/0000-0002-5386-465X","observed_orcids":["https://orcid.org/0000-0002-5386-465X"]},"institutions":[{"id":"https://openalex.org/I1291425158","display_name":"Google (United States)","ror":"https://ror.org/00njsd438","country_code":"US","type":"company","lineage":["https://openalex.org/I1291425158","https://openalex.org/I4210128969"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Pietquin, Olivier","raw_affiliation_strings":["Google Research , Brain Team","Google,,,,,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Research , Brain Team","institution_ids":["https://openalex.org/I1291425158"]},{"raw_affiliation_string":"Google,,,,,","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5006533777","display_name":"R\u00e9mi Munos","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I1291425158","display_name":"Google (United States)","ror":"https://ror.org/00njsd438","country_code":"US","type":"company","lineage":["https://openalex.org/I1291425158","https://openalex.org/I4210128969"]},{"id":"https://openalex.org/I1294671590","display_name":"Centre National de la Recherche Scientifique","ror":"https://ror.org/02feahw73","country_code":"FR","type":"government","lineage":["https://openalex.org/I1294671590"]},{"id":"https://openalex.org/I3017839442","display_name":"Institut \u00c9lie Cartan de Lorraine","ror":"https://ror.org/04rvw8791","country_code":"FR","type":"facility","lineage":["https://openalex.org/I1294671590","https://openalex.org/I1326498283","https://openalex.org/I3017839442","https://openalex.org/I4210141950","https://openalex.org/I90183372"]},{"id":"https://openalex.org/I90183372","display_name":"Universit\u00e9 de Lorraine","ror":"https://ror.org/04vfs2w97","country_code":"FR","type":"education","lineage":["https://openalex.org/I90183372"]}],"countries":["FR","US"],"is_corresponding":false,"raw_author_name":"Munos, R\u00e9mi","raw_affiliation_strings":["Universit\u00e9 de Lorraine , CNRS , Inria IECL , F-54000 Nancy , France","Google,,,,,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Universit\u00e9 de Lorraine , CNRS , Inria IECL , F-54000 Nancy , France","institution_ids":["https://openalex.org/I1294671590","https://openalex.org/I3017839442","https://openalex.org/I90183372"]},{"raw_affiliation_string":"Google,,,,,","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5110482875","display_name":"Matthieu Geist","orcid":"https://orcid.org/0009-0002-7888-8233","observed_orcids":["https://orcid.org/0009-0002-7888-8233"]},"institutions":[{"id":"https://openalex.org/I1291425158","display_name":"Google (United States)","ror":"https://ror.org/00njsd438","country_code":"US","type":"company","lineage":["https://openalex.org/I1291425158","https://openalex.org/I4210128969"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Geist, Matthieu","raw_affiliation_strings":["Google Research , Brain Team","Vancouver , Canada","Google,,,,,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Research , Brain Team","institution_ids":["https://openalex.org/I1291425158"]},{"raw_affiliation_string":"Vancouver , Canada","institution_ids":[]},{"raw_affiliation_string":"Google,,,,,","institution_ids":["https://openalex.org/I1291425158"]}]}],"institutions":[],"countries_distinct_count":3,"institutions_distinct_count":6,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":22,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10462","display_name":"Reinforcement Learning in Robotics","score":1.0,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10462","display_name":"Reinforcement Learning in Robotics","score":1.0,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T12101","display_name":"Advanced Bandit Algorithms Research","score":0.0,"subfield":{"id":"https://openalex.org/subfields/1803","display_name":"Management Science and Operations Research"},"field":{"id":"https://openalex.org/fields/18","display_name":"Decision Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T12072","display_name":"Machine Learning and Algorithms","score":0.0,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/entropy-regularization","display_name":"entropy regularization","score":0.7440000176429749},{"id":"https://openalex.org/keywords/approximate-value-iteration","display_name":"approximate value iteration","score":0.878000020980835},{"id":"https://openalex.org/keywords/performance-bounds","display_name":"performance bounds","score":0.7490000128746033},{"id":"https://openalex.org/keywords/error-propagation","display_name":"error propagation","score":0.6420000195503235}],"concepts":[{"id":"https://openalex.org/C2776135515","wikidata":"https://www.wikidata.org/wiki/Q17143721","display_name":"Regularization (linguistics)","level":2,"score":0.6978570818901062},{"id":"https://openalex.org/C129844170","wikidata":"https://www.wikidata.org/wiki/Q41299","display_name":"Quadratic equation","level":2,"score":0.6315973997116089},{"id":"https://openalex.org/C153083717","wikidata":"https://www.wikidata.org/wiki/Q6535263","display_name":"Leverage (statistics)","level":2,"score":0.5902001261711121},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.5888667106628418},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5380463600158691},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.48146477341651917},{"id":"https://openalex.org/C28826006","wikidata":"https://www.wikidata.org/wiki/Q33521","display_name":"Applied mathematics","level":1,"score":0.45999419689178467},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4365774691104889},{"id":"https://openalex.org/C106301342","wikidata":"https://www.wikidata.org/wiki/Q4117933","display_name":"Entropy (arrow of time)","level":2,"score":0.4294361472129822},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.37787219882011414},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.2695506811141968},{"id":"https://openalex.org/C62520636","wikidata":"https://www.wikidata.org/wiki/Q944","display_name":"Quantum mechanics","level":1,"score":0.0},{"id":"https://openalex.org/C2524010","wikidata":"https://www.wikidata.org/wiki/Q8087","display_name":"Geometry","level":1,"score":0.0},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.0}],"mesh":[],"locations_count":3,"locations":[{"id":"pmh:oai:arXiv.org:2003.14089","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2003.14089","pdf_url":"https://arxiv.org/pdf/2003.14089","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"mag:3014137283","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/pdf/2003.14089.pdf","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"arXiv (Cornell University)","raw_type":null},{"id":"doi:10.48550/arxiv.2003.14089","doi":"https://doi.org/10.48550/arxiv.2003.14089","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2003.14089","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2003.14089","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2003.14089","pdf_url":"https://arxiv.org/pdf/2003.14089","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W3014137283.pdf","grobid_xml":"https://content.openalex.org/works/W3014137283.grobid-xml"},"referenced_works_count":42,"referenced_works":["https://openalex.org/W1503425191","https://openalex.org/W1575592356","https://openalex.org/W1771410628","https://openalex.org/W1788877992","https://openalex.org/W1850731992","https://openalex.org/W1889629917","https://openalex.org/W1994123361","https://openalex.org/W2020609518","https://openalex.org/W2072931156","https://openalex.org/W2100100236","https://openalex.org/W2119567691","https://openalex.org/W2145339207","https://openalex.org/W2149166950","https://openalex.org/W2150459019","https://openalex.org/W2150468603","https://openalex.org/W2165421048","https://openalex.org/W2260756217","https://openalex.org/W2411690432","https://openalex.org/W2462023919","https://openalex.org/W2594103415","https://openalex.org/W2609650878","https://openalex.org/W2736601468","https://openalex.org/W2751651528","https://openalex.org/W2781726626","https://openalex.org/W2785738552","https://openalex.org/W2799151646","https://openalex.org/W2898757087","https://openalex.org/W2903383131","https://openalex.org/W2905342215","https://openalex.org/W2914920107","https://openalex.org/W2921114252","https://openalex.org/W2945159000","https://openalex.org/W2945496654","https://openalex.org/W2962847657","https://openalex.org/W2963169817","https://openalex.org/W2963267001","https://openalex.org/W2963923407","https://openalex.org/W2981073936","https://openalex.org/W2998050631","https://openalex.org/W3099050578","https://openalex.org/W4300428759","https://openalex.org/W6990719774"],"related_works":["https://openalex.org/W4287822321","https://openalex.org/W3101487584","https://openalex.org/W3183475735","https://openalex.org/W7146166193","https://openalex.org/W3193494395","https://openalex.org/W4407424870","https://openalex.org/W4404396927","https://openalex.org/W4406603692","https://openalex.org/W4414817620","https://openalex.org/W4415274511"],"abstract_inverted_index":{"Recent":[0],"Reinforcement":[1],"Learning":[2],"(RL)":[3],"algorithms":[4],"making":[5],"use":[6],"of":[7,74,85,90,101,116],"Kullback-Leibler":[8],"(KL)":[9],"regularization":[10,28,35],"as":[11],"a":[12,54,67],"core":[13],"component":[14],"have":[15],"shown":[16],"outstanding":[17],"performance.":[18],"Yet,":[19],"only":[20],"little":[21],"is":[22],"understood":[23],"theoretically":[24],"about":[25],"why":[26],"KL":[27,34],"helps,":[29],"so":[30,125],"far.":[31],"We":[32,94],"study":[33,96],"within":[36],"an":[37,77,82,91,102,132],"approximate":[38],"value":[39],"iteration":[40],"scheme":[41,109],"and":[42,76],"show":[43],"that":[44],"it":[45],"implicitly":[46],"averages":[47],"q-values.":[48],"Leveraging":[49],"this":[50,128],"insight,":[51],"we":[52,126],"provide":[53],"very":[55,60],"strong":[56],"performance":[57],"bound,":[58],"the":[59,71,86,97],"first":[61],"to":[62,70],"combine":[63],"two":[64],"desirable":[65],"aspects:":[66],"linear":[68],"dependency":[69],"horizon":[72],"(instead":[73,89],"quadratic)":[75],"error":[78],"propagation":[79],"term":[80],"involving":[81],"averaging":[83],"effect":[84],"estimation":[87],"errors":[88],"accumulation":[92],"effect).":[93],"also":[95],"more":[98],"general":[99],"case":[100],"additional":[103],"entropy":[104],"regularizer.":[105],"The":[106],"resulting":[107],"abstract":[108],"encompasses":[110],"many":[111],"existing":[112],"RL":[113],"algorithms.":[114],"Some":[115],"our":[117],"assumptions":[118],"do":[119],"not":[120],"hold":[121],"with":[122,131],"neural":[123],"networks,":[124],"complement":[127],"theoretical":[129],"analysis":[130],"extensive":[133],"empirical":[134],"study.":[135]},"counts_by_year":[{"year":2026,"cited_by_count":1},{"year":2023,"cited_by_count":1},{"year":2022,"cited_by_count":3},{"year":2021,"cited_by_count":11},{"year":2020,"cited_by_count":6}],"updated_date":"2026-11-05T12:25:35.769452","created_date":"2020-04-10T00:00:00"}