{"id":"https://openalex.org/W7160561377","doi":"https://doi.org/10.48550/arxiv.2605.04065","title":"Free Energy-Driven Reinforcement Learning with Adaptive Advantage Shaping for Unsupervised Reasoning in LLMs","display_name":"Free Energy-Driven Reinforcement Learning with Adaptive Advantage Shaping for Unsupervised Reasoning in LLMs","publication_year":2026,"publication_date":"2026-04-11","ids":{"openalex":"https://openalex.org/W7160561377","doi":"https://doi.org/10.48550/arxiv.2605.04065"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.04065","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.04065","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"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":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.04065","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5135592276","display_name":"Yiming Huang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Yiming","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135541939","display_name":"Zhenbo Shi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shi, Zhenbo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5047620514","display_name":"Xin-Cheng Wen","orcid":"https://orcid.org/0000-0002-2115-9921"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wen, Xin-Cheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5024766544","display_name":"Jichuan Zeng","orcid":"https://orcid.org/0000-0003-4073-1214"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zeng, Jichuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135571879","display_name":"Cuiyun Gao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gao, Cuiyun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135616272","display_name":"Peiyi Han","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Han, Peiyi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5135625209","display_name":"Chuanyi Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Chuanyi","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":null,"has_fulltext":false,"cited_by_count":0,"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":0.2928999960422516,"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":0.2928999960422516,"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":0.1264999955892563,"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":0.04969999939203262,"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":0.667900025844574},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.560699999332428},{"id":"https://openalex.org/keywords/unsupervised-learning","display_name":"Unsupervised learning","score":0.5357999801635742},{"id":"https://openalex.org/keywords/process","display_name":"Process (computing)","score":0.34310001134872437},{"id":"https://openalex.org/keywords/empirical-research","display_name":"Empirical research","score":0.3192000091075897},{"id":"https://openalex.org/keywords/statistical-learning","display_name":"Statistical learning","score":0.2987000048160553}],"concepts":[{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.7017999887466431},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6912000179290771},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.667900025844574},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.5906999707221985},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.560699999332428},{"id":"https://openalex.org/C8038995","wikidata":"https://www.wikidata.org/wiki/Q1152135","display_name":"Unsupervised learning","level":2,"score":0.5357999801635742},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.34310001134872437},{"id":"https://openalex.org/C120936955","wikidata":"https://www.wikidata.org/wiki/Q2155640","display_name":"Empirical research","level":2,"score":0.3192000091075897},{"id":"https://openalex.org/C2982736386","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Statistical learning","level":2,"score":0.2987000048160553},{"id":"https://openalex.org/C125014702","wikidata":"https://www.wikidata.org/wiki/Q4680749","display_name":"Adaptive learning","level":2,"score":0.27950000762939453},{"id":"https://openalex.org/C79416737","wikidata":"https://www.wikidata.org/wiki/Q2305519","display_name":"Social learning","level":2,"score":0.27129998803138733},{"id":"https://openalex.org/C166052673","wikidata":"https://www.wikidata.org/wiki/Q83021","display_name":"Empirical evidence","level":2,"score":0.26739999651908875},{"id":"https://openalex.org/C186370098","wikidata":"https://www.wikidata.org/wiki/Q442787","display_name":"Energy (signal processing)","level":2,"score":0.25110000371932983}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.04065","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.04065","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"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:10.48550/arxiv.2605.04065","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.04065","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"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":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/7","score":0.8385677337646484,"display_name":"Affordable and clean energy"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Unsupervised":[0],"reinforcement":[1],"learning":[2,91],"(RL)":[3],"has":[4],"emerged":[5],"as":[6],"a":[7,57],"promising":[8],"paradigm":[9],"for":[10],"enabling":[11],"self-improvement":[12],"in":[13,44,119,135],"large":[14],"language":[15],"models":[16],"(LLMs).":[17],"However,":[18],"existing":[19],"unsupervised":[20,115],"RL-based":[21,59,116],"methods":[22,39,126],"often":[23],"lack":[24],"the":[25,30,45,80,95,138],"capacity":[26],"to":[27,29,73,132],"adapt":[28],"model's":[31],"evolving":[32],"reasoning":[33,108,121],"capabilities":[34],"during":[35],"training.":[36],"Therefore,":[37],"these":[38],"can":[40],"misdirect":[41],"policy":[42],"optimization":[43],"absence":[46],"of":[47,98,130],"ground-truth":[48],"supervision.":[49],"To":[50],"address":[51],"this":[52],"issue,":[53],"we":[54],"introduce":[55],"FREIA,":[56],"novel":[58],"algorithm":[60],"built":[61],"on":[62,79,94,103],"two":[63],"key":[64],"innovations:":[65],"(1)":[66],"Free":[67,81],"Energy-Driven":[68],"Reward":[69],"(FER)":[70],"adapts":[71],"rewards":[72],"balance":[74],"consensus":[75],"and":[76],"exploration":[77],"based":[78,93],"Energy":[82],"Principle.":[83],"(2)":[84],"Adaptive":[85],"Advantage":[86],"Shaping":[87],"(AAS)":[88],"adaptively":[89],"adjusts":[90],"signals":[92],"statistical":[96],"characteristics":[97],"sampled":[99],"rewards.":[100],"Empirical":[101],"evaluations":[102],"nine":[104],"datasets":[105],"across":[106],"three":[107],"tasks":[109],"showcase":[110],"that":[111],"FREIA":[112,123],"outperforms":[113],"other":[114,125],"baselines.":[117],"Notably,":[118],"mathematical":[120],"tasks,":[122],"surpasses":[124],"by":[127],"an":[128],"average":[129],"0.5":[131],"3.5":[133],"points":[134],"Pass@1":[136],"using":[137],"DeepSeek-R1-Distill-Qwen-1.5B":[139],"model.":[140]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-08T00:00:00"}