{"id":"https://openalex.org/W4414821027","doi":"https://doi.org/10.48550/arxiv.2510.02172","title":"RESTRAIN: From Spurious Votes to Signals -- Self-Driven RL with Self-Penalization","display_name":"RESTRAIN: From Spurious Votes to Signals -- Self-Driven RL with Self-Penalization","publication_year":2025,"publication_date":"2025-10-02","ids":{"openalex":"https://openalex.org/W4414821027","doi":"https://doi.org/10.48550/arxiv.2510.02172"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2510.02172","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2510.02172","pdf_url":"https://arxiv.org/pdf/2510.02172","source":{"id":"https://openalex.org/S4393918464","display_name":"ArXiv.org","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"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"},"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2510.02172","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5015497573","display_name":"Zhongjie Yu","orcid":"https://orcid.org/0000-0002-4935-0154"},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Yu, Zhaoning","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5085821540","display_name":"Wanghui Su","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Su, Will","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5119845742","display_name":"Leitian Tao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tao, Leitian","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Wang, Haozhu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Haozhu","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Singh, Aashu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Singh, Aashu","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5028988765","display_name":"Hanchao Yu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yu, Hanchao","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100398965","display_name":"Jianyu Wang","orcid":"https://orcid.org/0000-0002-2180-6033"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Jianyu","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Gao, Hongyang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gao, Hongyang","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5114159551","display_name":"Weizhe Yuan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yuan, Weizhe","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5119845741","display_name":"Jason Weston","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Weston, Jason","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101486490","display_name":"Ping Yu","orcid":"https://orcid.org/0000-0002-8505-8898"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yu, Ping","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5114860140","display_name":"Jing Xu","orcid":"https://orcid.org/0009-0001-3504-691X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xu, Jing","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":12,"corresponding_author_ids":["https://openalex.org/A5015497573"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"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/T10270","display_name":"Blockchain Technology Applications and Security","score":0.6682000160217285,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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/T10270","display_name":"Blockchain Technology Applications and Security","score":0.6682000160217285,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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/T11598","display_name":"Internet Traffic Analysis and Secure E-voting","score":0.5902000069618225,"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/spurious-relationship","display_name":"Spurious relationship","score":0.8187999725341797},{"id":"https://openalex.org/keywords/scalability","display_name":"Scalability","score":0.6812000274658203},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.6676999926567078},{"id":"https://openalex.org/keywords/matching","display_name":"Matching (statistics)","score":0.5733000040054321},{"id":"https://openalex.org/keywords/exploit","display_name":"Exploit","score":0.49050000309944153},{"id":"https://openalex.org/keywords/path","display_name":"Path (computing)","score":0.4440999925136566},{"id":"https://openalex.org/keywords/training-set","display_name":"Training set","score":0.37139999866485596}],"concepts":[{"id":"https://openalex.org/C97256817","wikidata":"https://www.wikidata.org/wiki/Q1462316","display_name":"Spurious relationship","level":2,"score":0.8187999725341797},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7091000080108643},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.6812000274658203},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.6676999926567078},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5885000228881836},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.5733000040054321},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.49050000309944153},{"id":"https://openalex.org/C2777735758","wikidata":"https://www.wikidata.org/wiki/Q817765","display_name":"Path (computing)","level":2,"score":0.4440999925136566},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.43700000643730164},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.37139999866485596},{"id":"https://openalex.org/C2777868144","wikidata":"https://www.wikidata.org/wiki/Q7239817","display_name":"Preference elicitation","level":3,"score":0.3637000024318695},{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.3215999901294708},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.31790000200271606},{"id":"https://openalex.org/C2776608160","wikidata":"https://www.wikidata.org/wiki/Q4785462","display_name":"Natural (archaeology)","level":2,"score":0.26409998536109924},{"id":"https://openalex.org/C77967617","wikidata":"https://www.wikidata.org/wiki/Q4677561","display_name":"Active learning (machine learning)","level":2,"score":0.2558000087738037},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.2529999911785126},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.2513999938964844}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2510.02172","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2510.02172","pdf_url":"https://arxiv.org/pdf/2510.02172","source":{"id":"https://openalex.org/S4393918464","display_name":"ArXiv.org","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"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.2510.02172","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2510.02172","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":"article"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2510.02172","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2510.02172","pdf_url":"https://arxiv.org/pdf/2510.02172","source":{"id":"https://openalex.org/S4393918464","display_name":"ArXiv.org","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"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"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4414821027.pdf"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Reinforcement":[0],"learning":[1,50,67],"with":[2,51],"human-annotated":[3],"data":[4,22],"has":[5],"boosted":[6],"chain-of-thought":[7],"reasoning":[8,11,94,115,169],"in":[9,20],"large":[10,119],"models,":[12],"but":[13],"these":[14],"gains":[15,120],"come":[16],"at":[17],"high":[18],"costs":[19],"labeled":[21],"while":[23,91,153],"faltering":[24],"on":[25,138,142,147],"harder":[26],"tasks.":[27],"A":[28],"natural":[29],"next":[30],"step":[31],"is":[32],"experience-driven":[33],"learning,":[34],"where":[35],"models":[36],"improve":[37],"without":[38,111,170],"curated":[39],"labels":[40,63],"by":[41,133],"adapting":[42],"to":[43,72,135],"unlabeled":[44,123],"data.":[45,124],"We":[46],"introduce":[47],"RESTRAIN":[48,76,117,162],"(REinforcement":[49],"Self-restraint),":[52],"a":[53,65,164],"self-penalizing":[54],"RL":[55],"framework":[56],"that":[57,161],"converts":[58],"the":[59,80],"absence":[60],"of":[61,70],"gold":[62,156,171],"into":[64,101],"useful":[66],"signal.":[68],"Instead":[69],"overcommitting":[71],"spurious":[73],"majority":[74],"votes,":[75],"exploits":[77],"signals":[78],"from":[79],"model's":[81],"entire":[82],"answer":[83],"distribution:":[84],"penalizing":[85],"overconfident":[86],"rollouts":[87],"and":[88,127,144],"low-consistency":[89],"examples":[90],"preserving":[92],"promising":[93],"chains.":[95],"The":[96],"self-penalization":[97],"mechanism":[98],"integrates":[99],"seamlessly":[100],"policy":[102],"optimization":[103],"methods":[104],"such":[105],"as":[106],"GRPO,":[107],"enabling":[108],"continual":[109],"self-improvement":[110],"supervision.":[112],"On":[113],"challenging":[114],"benchmarks,":[116],"delivers":[118],"using":[121,154],"only":[122],"With":[125],"Qwen3-4B-Base":[126],"OctoThinker":[128],"Hybrid-8B-Base,":[129],"it":[130],"improves":[131],"Pass@1":[132],"up":[134],"+140.7":[136],"percent":[137,141,146],"AIME25,":[139],"+36.2":[140],"MMLU_STEM,":[143],"+19.6":[145],"GPQA-Diamond,":[148],"nearly":[149],"matching":[150],"gold-label":[151],"training":[152],"no":[155],"labels.":[157,172],"These":[158],"results":[159],"demonstrate":[160],"establishes":[163],"scalable":[165],"path":[166],"toward":[167],"stronger":[168]},"counts_by_year":[],"updated_date":"2026-04-21T08:09:41.155169","created_date":"2025-10-10T00:00:00"}