{"id":"https://openalex.org/W4415037569","doi":"https://doi.org/10.48550/arxiv.2505.21444","title":"Can Large Reasoning Models Self-Train?","display_name":"Can Large Reasoning Models Self-Train?","publication_year":2025,"publication_date":"2025-05-27","ids":{"openalex":"https://openalex.org/W4415037569","doi":"https://doi.org/10.48550/arxiv.2505.21444"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2505.21444","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2505.21444","pdf_url":"https://arxiv.org/pdf/2505.21444","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":"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"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2505.21444","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5093766597","display_name":"Sheikh Shafayat","orcid":"https://orcid.org/0009-0000-7823-2133","observed_orcids":["https://orcid.org/0009-0000-7823-2133"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shafayat, Sheikh","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5046483574","display_name":"Fahim Tajwar","orcid":"https://orcid.org/0000-0001-9257-6282","observed_orcids":["https://orcid.org/0000-0001-9257-6282"]},"institutions":[{"id":"https://openalex.org/I74973139","display_name":"Carnegie Mellon University","ror":"https://ror.org/05x2bcf33","country_code":"US","type":"education","lineage":["https://openalex.org/I74973139"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Tajwar, Fahim","raw_affiliation_strings":["Carnegie Mellon University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Carnegie Mellon University","institution_ids":["https://openalex.org/I74973139"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5071983998","display_name":"Ruslan Salakhutdinov","orcid":"https://orcid.org/0000-0002-3752-2756","observed_orcids":["https://orcid.org/0000-0002-3752-2756"]},"institutions":[{"id":"https://openalex.org/I74973139","display_name":"Carnegie Mellon University","ror":"https://ror.org/05x2bcf33","country_code":"US","type":"education","lineage":["https://openalex.org/I74973139"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Salakhutdinov, Ruslan","raw_affiliation_strings":["Carnegie Mellon University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Carnegie Mellon University","institution_ids":["https://openalex.org/I74973139"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5055199976","display_name":"Jeff Schneider","orcid":"https://orcid.org/0000-0002-5080-9073","observed_orcids":["https://orcid.org/0000-0002-5080-9073"]},"institutions":[{"id":"https://openalex.org/I74973139","display_name":"Carnegie Mellon University","ror":"https://ror.org/05x2bcf33","country_code":"US","type":"education","lineage":["https://openalex.org/I74973139"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Schneider, Jeff","raw_affiliation_strings":["Carnegie Mellon University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Carnegie Mellon University","institution_ids":["https://openalex.org/I74973139"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5055456006","display_name":"Andrea Zanette","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I74973139","display_name":"Carnegie Mellon University","ror":"https://ror.org/05x2bcf33","country_code":"US","type":"education","lineage":["https://openalex.org/I74973139"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zanette, Andrea","raw_affiliation_strings":["Carnegie Mellon University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Carnegie Mellon University","institution_ids":["https://openalex.org/I74973139"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"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/T10462","display_name":"Reinforcement Learning in Robotics","score":0.8041999936103821,"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":0.8041999936103821,"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/T12072","display_name":"Machine Learning and Algorithms","score":0.05460000038146973,"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":0.02280000038444996,"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/self-training","display_name":"self-training","score":0.949999988079071},{"id":"https://openalex.org/keywords/large-reasoning-models","display_name":"large reasoning models","score":0.7590000033378601},{"id":"https://openalex.org/keywords/majority-voting","display_name":"majority voting","score":0.7440000176429749},{"id":"https://openalex.org/keywords/self-feedback","display_name":"self-feedback","score":0.7799999713897705},{"id":"https://openalex.org/keywords/reward-hacking","display_name":"reward hacking","score":0.7070000171661377},{"id":"https://openalex.org/keywords/personal-development","display_name":"human development","score":0.6690000295639038},{"id":"https://openalex.org/keywords/feedback-control-design","display_name":"feedback control design","score":0.8240000009536743},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"reinforcement learning","score":0.4429999887943268},{"id":"https://openalex.org/keywords/reasoning-performance","display_name":"reasoning performance","score":0.4180000126361847}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7371000051498413},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.6107000112533569},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.5960000157356262},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.5705000162124634},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5673999786376953},{"id":"https://openalex.org/C2780586882","wikidata":"https://www.wikidata.org/wiki/Q7520643","display_name":"Simple (philosophy)","level":2,"score":0.45579999685287476},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.4510999917984009},{"id":"https://openalex.org/C520049643","wikidata":"https://www.wikidata.org/wiki/Q189760","display_name":"Voting","level":3,"score":0.397599995136261},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.38760000467300415},{"id":"https://openalex.org/C37335422","wikidata":"https://www.wikidata.org/wiki/Q6888134","display_name":"Model-based reasoning","level":3,"score":0.31459999084472656},{"id":"https://openalex.org/C20162079","wikidata":"https://www.wikidata.org/wiki/Q1151406","display_name":"Case-based reasoning","level":2,"score":0.2897000014781952},{"id":"https://openalex.org/C159032336","wikidata":"https://www.wikidata.org/wiki/Q2488768","display_name":"Non-monotonic logic","level":2,"score":0.2612999975681305},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.2606000006198883}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2505.21444","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2505.21444","pdf_url":"https://arxiv.org/pdf/2505.21444","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":"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.2505.21444","doi":"https://doi.org/10.48550/arxiv.2505.21444","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2505.21444","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":"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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2505.21444","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2505.21444","pdf_url":"https://arxiv.org/pdf/2505.21444","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":"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":[],"study_designs":[],"awards":[{"id":"https://openalex.org/G4804575867","display_name":"Category I: Crossing the Divide Between Today's Practice and Tomorrow's Science","funder_award_id":"2005572","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"}],"funders":[{"id":"https://openalex.org/F4320306076","display_name":"National Science Foundation","ror":"https://ror.org/021nxhr62"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4415037569.pdf","grobid_xml":"https://content.openalex.org/works/W4415037569.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4416365886","https://openalex.org/W4416550548","https://openalex.org/W4415967455","https://openalex.org/W7162432771","https://openalex.org/W4417257705","https://openalex.org/W7131777951","https://openalex.org/W7119699747","https://openalex.org/W7204940568","https://openalex.org/W7203490308","https://openalex.org/W6967252882"],"abstract_inverted_index":{"Recent":[0],"successes":[1],"of":[2,14,53,79,101],"reinforcement":[3],"learning":[4],"(RL)":[5],"in":[6,123],"training":[7,120],"large":[8],"reasoning":[9,60,73],"models":[10,116],"motivate":[11],"the":[12,18,71,85,136],"question":[13,40],"whether":[15],"self-training":[16,104],"-":[17,28,106],"process":[19],"where":[20,115],"a":[21,45,50,98,103],"model":[22,91],"learns":[23],"from":[24],"its":[25,77],"own":[26],"judgments":[27],"can":[29],"be":[30],"sustained":[31],"within":[32],"RL.":[33],"In":[34],"this":[35,39,65],"work,":[36],"we":[37,62],"study":[38],"using":[41],"majority":[42],"voting":[43],"as":[44,135],"simple":[46],"self-feedback":[47],"mechanism.":[48],"On":[49],"comprehensive":[51],"set":[52],"experiments":[54],"on":[55,144],"both":[56],"synthetic":[57],"and":[58,125,139],"real":[59],"tasks,":[61],"find":[63],"that":[64],"basic":[66],"approach":[67],"improves":[68],"not":[69],"only":[70],"model's":[72],"performance,":[74],"but":[75],"also":[76,96],"capability":[78],"generating":[80],"better":[81],"quality":[82],"feedback":[83,133],"for":[84,141],"next":[86],"RL":[87,108],"iteration,":[88],"driving":[89],"further":[90],"improvement.":[92],"Yet":[93],"our":[94],"analysis":[95],"reveals":[97],"critical":[99],"limitation":[100],"such":[102],"paradigm":[105],"prolonged":[107,148],"with":[109],"self-reward":[110],"leads":[111],"to":[112,118,146],"reward":[113],"hacking":[114],"learn":[117],"maximize":[119],"(pseudo-)reward,":[121],"resulting":[122],"sudden":[124],"complete":[126],"performance":[127],"collapse.":[128],"Together,":[129],"these":[130],"results":[131],"highlight":[132],"design":[134],"central":[137],"challenge":[138],"call":[140],"future":[142],"research":[143],"mechanisms":[145],"enable":[147],"self-improvement.":[149]},"counts_by_year":[],"updated_date":"2026-09-30T09:27:27.532063","created_date":"2025-10-10T00:00:00"}