{"id":"https://openalex.org/W7138762739","doi":"https://doi.org/10.1609/aaai.v40i37.40442","title":"ENCORE: Entropy-guided Reward Composition for Multi-head Safety Reward Models","display_name":"ENCORE: Entropy-guided Reward Composition for Multi-head Safety Reward Models","publication_year":2026,"publication_date":"2026-03-14","ids":{"openalex":"https://openalex.org/W7138762739","doi":"https://doi.org/10.1609/aaai.v40i37.40442"},"language":null,"primary_location":{"id":"doi:10.1609/aaai.v40i37.40442","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i37.40442","pdf_url":null,"source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"diamond","oa_url":"https://doi.org/10.1609/aaai.v40i37.40442","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5129745659","display_name":"Xiaomin Li","orcid":null},"institutions":[{"id":"https://openalex.org/I2801851002","display_name":"Harvard University Press","ror":"https://ror.org/006v7bf86","country_code":"US","type":"other","lineage":["https://openalex.org/I136199984","https://openalex.org/I2801851002"]}],"countries":["US"],"is_corresponding":true,"raw_author_name":"Xiaomin Li","raw_affiliation_strings":["Harvard University"],"affiliations":[{"raw_affiliation_string":"Harvard University","institution_ids":["https://openalex.org/I2801851002"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129673977","display_name":"Xupeng Chen","orcid":null},"institutions":[{"id":"https://openalex.org/I57206974","display_name":"New York University","ror":"https://ror.org/0190ak572","country_code":"US","type":"education","lineage":["https://openalex.org/I57206974"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Xupeng Chen","raw_affiliation_strings":["New York University"],"affiliations":[{"raw_affiliation_string":"New York University","institution_ids":["https://openalex.org/I57206974"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129674129","display_name":"Jingxuan Fan","orcid":null},"institutions":[{"id":"https://openalex.org/I2801851002","display_name":"Harvard University Press","ror":"https://ror.org/006v7bf86","country_code":"US","type":"other","lineage":["https://openalex.org/I136199984","https://openalex.org/I2801851002"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Jingxuan Fan","raw_affiliation_strings":["Harvard University"],"affiliations":[{"raw_affiliation_string":"Harvard University","institution_ids":["https://openalex.org/I2801851002"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5088789651","display_name":"Eric Hanchen Jiang","orcid":null},"institutions":[{"id":"https://openalex.org/I161318765","display_name":"University of California, Los Angeles","ror":"https://ror.org/046rm7j60","country_code":"US","type":"education","lineage":["https://openalex.org/I161318765"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Eric Hanchen Jiang","raw_affiliation_strings":["University of California, Los Angeles"],"affiliations":[{"raw_affiliation_string":"University of California, Los Angeles","institution_ids":["https://openalex.org/I161318765"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5129679786","display_name":"Mingye Gao","orcid":null},"institutions":[{"id":"https://openalex.org/I63966007","display_name":"Massachusetts Institute of Technology","ror":"https://ror.org/042nb2s44","country_code":"US","type":"education","lineage":["https://openalex.org/I63966007"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Mingye Gao","raw_affiliation_strings":["Massachusetts Institute of Technology"],"affiliations":[{"raw_affiliation_string":"Massachusetts Institute of Technology","institution_ids":["https://openalex.org/I63966007"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":5,"corresponding_author_ids":["https://openalex.org/A5129745659"],"corresponding_institution_ids":["https://openalex.org/I2801851002"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.70021413,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":"40","issue":"37","first_page":"31743","last_page":"31750"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.16910000145435333,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.16910000145435333,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.1242000013589859,"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":0.08089999854564667,"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","display_name":"Entropy (arrow of time)","score":0.6333000063896179},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.5401999950408936},{"id":"https://openalex.org/keywords/construct","display_name":"Construct (python library)","score":0.4415000081062317},{"id":"https://openalex.org/keywords/quality","display_name":"Quality (philosophy)","score":0.4383000135421753},{"id":"https://openalex.org/keywords/simple","display_name":"Simple (philosophy)","score":0.41690000891685486},{"id":"https://openalex.org/keywords/preference","display_name":"Preference","score":0.40939998626708984},{"id":"https://openalex.org/keywords/variety","display_name":"Variety (cybernetics)","score":0.319599986076355}],"concepts":[{"id":"https://openalex.org/C106301342","wikidata":"https://www.wikidata.org/wiki/Q4117933","display_name":"Entropy (arrow of time)","level":2,"score":0.6333000063896179},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6248999834060669},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.5401999950408936},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4837999939918518},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.47769999504089355},{"id":"https://openalex.org/C2780801425","wikidata":"https://www.wikidata.org/wiki/Q5164392","display_name":"Construct (python library)","level":2,"score":0.4415000081062317},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.4383000135421753},{"id":"https://openalex.org/C2780586882","wikidata":"https://www.wikidata.org/wiki/Q7520643","display_name":"Simple (philosophy)","level":2,"score":0.41690000891685486},{"id":"https://openalex.org/C2781249084","wikidata":"https://www.wikidata.org/wiki/Q908656","display_name":"Preference","level":2,"score":0.40939998626708984},{"id":"https://openalex.org/C136197465","wikidata":"https://www.wikidata.org/wiki/Q1729295","display_name":"Variety (cybernetics)","level":2,"score":0.319599986076355},{"id":"https://openalex.org/C134121241","wikidata":"https://www.wikidata.org/wiki/Q899301","display_name":"Yield (engineering)","level":2,"score":0.3122999966144562},{"id":"https://openalex.org/C9679016","wikidata":"https://www.wikidata.org/wiki/Q1417473","display_name":"Principle of maximum entropy","level":2,"score":0.3082999885082245},{"id":"https://openalex.org/C67203356","wikidata":"https://www.wikidata.org/wiki/Q1321905","display_name":"Reinforcement","level":2,"score":0.28119999170303345},{"id":"https://openalex.org/C181204326","wikidata":"https://www.wikidata.org/wiki/Q7239820","display_name":"Preference learning","level":3,"score":0.28029999136924744},{"id":"https://openalex.org/C83849319","wikidata":"https://www.wikidata.org/wiki/Q7295720","display_name":"Rating scale","level":2,"score":0.26249998807907104},{"id":"https://openalex.org/C40231798","wikidata":"https://www.wikidata.org/wiki/Q1333743","display_name":"Composition (language)","level":2,"score":0.2522999942302704}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1609/aaai.v40i37.40442","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i37.40442","pdf_url":null,"source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"}],"best_oa_location":{"id":"doi:10.1609/aaai.v40i37.40442","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i37.40442","pdf_url":null,"source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"sustainable_development_goals":[{"score":0.6176133751869202,"display_name":"Peace, Justice and strong institutions","id":"https://metadata.un.org/sdg/16"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"The":[0],"safety":[1,44,127],"alignment":[2],"of":[3,28],"large":[4],"language":[5],"models":[6],"(LLMs)":[7],"often":[8],"relies":[9],"on":[10,42,125],"reinforcement":[11],"learning":[12],"from":[13],"human":[14,19],"feedback":[15],"(RLHF),":[16],"which":[17],"requires":[18],"annotations":[20],"to":[21,33,60,78],"construct":[22],"preference":[23],"datasets.":[24],"Given":[25],"the":[26,99],"challenge":[27],"assigning":[29],"overall":[30],"quality":[31],"scores":[32],"data,":[34],"recent":[35],"works":[36],"increasingly":[37],"adopt":[38],"fine-grained":[39],"ratings":[40],"based":[41],"multiple":[43],"rules.":[45],"In":[46],"this":[47,69],"paper,":[48],"we":[49,71,90],"discover":[50],"a":[51,74,143],"robust":[52],"phenomenon:":[53],"Rules":[54],"with":[55,85],"higher":[56],"rating":[57,87],"entropy":[58],"tend":[59],"have":[61],"lower":[62],"accuracy":[63],"in":[64],"distinguishing":[65],"human-preferred":[66],"responses.":[67],"Exploiting":[68],"insight,":[70],"propose":[72],"ENCORE,":[73],"simple":[75],"entropy-guided":[76],"method":[77,130],"compose":[79],"multi-head":[80],"rewards":[81],"by":[82],"penalizing":[83],"rules":[84,94],"high":[86],"entropy.":[88],"Theoretically,":[89],"show":[91],"that":[92],"such":[93],"yield":[95],"negligible":[96],"weights":[97],"under":[98],"Bradley\u2013Terry":[100],"loss":[101],"during":[102],"weight":[103],"optimization,":[104],"naturally":[105],"justifying":[106],"their":[107],"penalization.":[108],"Empirically,":[109],"ENCORE":[110],"consistently":[111],"outperforms":[112],"strong":[113],"baselines,":[114],"including":[115],"random":[116],"and":[117,122,138,145],"uniform":[118],"weighting,":[119],"single-head":[120],"Bradley\u2013Terry,":[121],"LLM-as-a-judge,":[123],"etc.":[124],"RewardBench":[126],"tasks.":[128],"Our":[129],"is":[131],"completely":[132],"training-free,":[133],"generally":[134],"applicable":[135],"across":[136],"datasets,":[137],"retains":[139],"interpretability,":[140],"making":[141],"it":[142],"practical":[144],"effective":[146],"approach":[147],"for":[148],"multi-attribute":[149],"reward":[150],"modeling.":[151]},"counts_by_year":[],"updated_date":"2026-03-20T20:47:17.329874","created_date":"2026-03-18T00:00:00"}