{"id":"https://openalex.org/W4417143026","doi":"https://doi.org/10.18653/v1/2026.findings-acl.1559","title":"Entropy Ratio Clipping as a Soft Global Constraint for Stable Reinforcement Learning","display_name":"Entropy Ratio Clipping as a Soft Global Constraint for Stable Reinforcement Learning","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W4417143026","doi":"https://doi.org/10.18653/v1/2026.findings-acl.1559"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.findings-acl.1559","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.1559","pdf_url":"https://aclanthology.org/2026.findings-acl.1559.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.findings-acl.1559.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5081013230","display_name":"Zhenpeng Su","orcid":"https://orcid.org/0000-0001-5577-4538"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhenpeng Su","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101391648","display_name":"Leiyu Pan","orcid":"https://orcid.org/0009-0002-5304-0914"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Leiyu Pan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102026159","display_name":"Minghui Lv","orcid":"https://orcid.org/0000-0002-1871-8913"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Minxuan Lv","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5056504874","display_name":"T. Mei","orcid":"https://orcid.org/0000-0002-5646-9911"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tiehua Mei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Zijia Lin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zijia Lin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5006702189","display_name":"Yuntao Li","orcid":"https://orcid.org/0000-0002-8039-3240"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yuntao Li","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5103409760","display_name":"Wenping Hu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wenping Hu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5054330014","display_name":"Ruiming Tang","orcid":"https://orcid.org/0000-0002-9224-2431"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ruiming Tang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5062939922","display_name":"Kun Gai","orcid":"https://orcid.org/0000-0002-3636-3618"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kun Gai","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5108693980","display_name":"Guorui Zhou","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Guorui Zhou","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.03394277,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"31161","last_page":"31171"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10462","display_name":"Reinforcement Learning in Robotics","score":0.35199999809265137,"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":0.35199999809265137,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.11190000176429749,"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/T10028","display_name":"Topic Modeling","score":0.10540000349283218,"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/reinforcement-learning","display_name":"Reinforcement learning","score":0.7659000158309937},{"id":"https://openalex.org/keywords/entropy","display_name":"Entropy (arrow of time)","score":0.605400025844574},{"id":"https://openalex.org/keywords/kullback\u2013leibler-divergence","display_name":"Kullback\u2013Leibler divergence","score":0.4893999993801117},{"id":"https://openalex.org/keywords/performance-metric","display_name":"Performance metric","score":0.4611999988555908},{"id":"https://openalex.org/keywords/clipping","display_name":"Clipping (morphology)","score":0.4609000086784363},{"id":"https://openalex.org/keywords/probability-distribution","display_name":"Probability distribution","score":0.39559999108314514}],"concepts":[{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.7659000158309937},{"id":"https://openalex.org/C106301342","wikidata":"https://www.wikidata.org/wiki/Q4117933","display_name":"Entropy (arrow of time)","level":2,"score":0.605400025844574},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5914999842643738},{"id":"https://openalex.org/C171752962","wikidata":"https://www.wikidata.org/wiki/Q255166","display_name":"Kullback\u2013Leibler divergence","level":2,"score":0.4893999993801117},{"id":"https://openalex.org/C2780898871","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Performance metric","level":2,"score":0.4611999988555908},{"id":"https://openalex.org/C2776848632","wikidata":"https://www.wikidata.org/wiki/Q853463","display_name":"Clipping (morphology)","level":2,"score":0.4609000086784363},{"id":"https://openalex.org/C149441793","wikidata":"https://www.wikidata.org/wiki/Q200726","display_name":"Probability distribution","level":2,"score":0.39559999108314514},{"id":"https://openalex.org/C176217482","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Metric (unit)","level":2,"score":0.3944000005722046},{"id":"https://openalex.org/C2776036281","wikidata":"https://www.wikidata.org/wiki/Q48769818","display_name":"Constraint (computer-aided design)","level":2,"score":0.382999986410141},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.3668000102043152},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3564000129699707},{"id":"https://openalex.org/C9679016","wikidata":"https://www.wikidata.org/wiki/Q1417473","display_name":"Principle of maximum entropy","level":2,"score":0.31360000371932983},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.2818000018596649},{"id":"https://openalex.org/C112972136","wikidata":"https://www.wikidata.org/wiki/Q7595718","display_name":"Stability (learning theory)","level":2,"score":0.26660001277923584},{"id":"https://openalex.org/C167981619","wikidata":"https://www.wikidata.org/wiki/Q1685498","display_name":"Cross entropy","level":3,"score":0.263700008392334}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.18653/v1/2026.findings-acl.1559","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.1559","pdf_url":"https://aclanthology.org/2026.findings-acl.1559.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2512.05591","is_oa":true,"landing_page_url":"https://arxiv.org/abs/2512.05591","pdf_url":"https://arxiv.org/pdf/2512.05591","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,"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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2512.05591","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2512.05591","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,"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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.findings-acl.1559","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.1559","pdf_url":"https://aclanthology.org/2026.findings-acl.1559.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4417143026.pdf","grobid_xml":"https://content.openalex.org/works/W4417143026.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"language":[1],"model":[2,10],"post-training":[3],"relies":[4],"on":[5,91,106],"reinforcement":[6,139],"learning":[7,140],"to":[8,32,125],"improve":[9],"capability":[11],"and":[12,41,72,118,137],"alignment":[13],"quality.However,":[14],"the":[15,25,28,54,66,70,83,107,114,121],"offpolicy":[16],"training":[17,33],"paradigm":[18],"introduces":[19],"distribution":[20,116],"shift,":[21],"which":[22],"often":[23],"pushes":[24],"policy":[26,39,87,111],"beyond":[27],"trust":[29],"region,":[30],"leading":[31],"instabilities":[34],"manifested":[35],"as":[36,75],"fluctuations":[37],"in":[38,86],"entropy":[40,67,108],"unstable":[42],"gradients.Although":[43],"PPO-Clip":[44],"mitigates":[45],"this":[46,92],"issue":[47],"through":[48],"importance":[49],"clipping,":[50],"it":[51],"still":[52],"overlooks":[53],"global":[55,78,115],"distributional":[56],"shift":[57],"of":[58,123,129],"actions.To":[59],"address":[60],"these":[61],"challenges,":[62],"we":[63,94],"propose":[64],"using":[65],"ratio":[68],"between":[69],"current":[71],"previous":[73],"policies":[74],"a":[76],"new":[77],"metric":[79],"that":[80,102,146],"effectively":[81],"quantifies":[82],"relative":[84],"change":[85],"exploration":[88],"throughout":[89],"updates.Building":[90],"metric,":[93],"introduce":[95],"an":[96],"Entropy":[97],"Ratio":[98],"Clipping":[99],"(ERC)":[100],"mechanism":[101],"imposes":[103],"bidirectional":[104],"constraints":[105],"ratio.This":[109],"stabilizes":[110],"updates":[112],"at":[113],"level":[117],"compensates":[119],"for":[120],"inability":[122],"PPO-clip":[124],"regulate":[126],"probability":[127],"shifts":[128],"un-sampled":[130],"actions.We":[131],"integrate":[132],"ERC":[133,147],"into":[134],"both":[135],"DAPO":[136],"GPPO":[138],"algorithms.Experiments":[141],"across":[142],"multiple":[143],"benchmarks":[144],"show":[145],"consistently":[148],"improves":[149],"performance.":[150]},"counts_by_year":[],"updated_date":"2026-08-01T09:00:35.917206","created_date":"2025-12-09T00:00:00"}