{"id":"https://openalex.org/W7161600156","doi":"https://doi.org/10.48550/arxiv.2605.15224","title":"ICRL: Learning to Internalize Self-Critique with Reinforcement Learning","display_name":"ICRL: Learning to Internalize Self-Critique with Reinforcement Learning","publication_year":2026,"publication_date":"2026-05-13","ids":{"openalex":"https://openalex.org/W7161600156","doi":"https://doi.org/10.48550/arxiv.2605.15224"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.15224","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.15224","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":null,"license_id":null,"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.2605.15224","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5136414422","display_name":"Jianbo Lin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lin, Jianbo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136384260","display_name":"Xiaomin Yu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yu, Xiaomin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136359752","display_name":"Yi Xin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xin, Yi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5114087923","display_name":"Yifu Guo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Guo, Yifu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136370777","display_name":"Zhuosong Jiang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jiang, Zhuosong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5127834758","display_name":"Zhongqi Yue","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yue, Zhongqi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136384527","display_name":"Weishi Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Weishi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5023152132","display_name":"Heqing Zou","orcid":"https://orcid.org/0000-0003-0038-2822"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zou, Heqing","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136434936","display_name":"Chengwei Qin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qin, Chengwei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5136390291","display_name":"Hui Xiong","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiong, Hui","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/T10028","display_name":"Topic Modeling","score":0.35899999737739563,"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/T10028","display_name":"Topic Modeling","score":0.35899999737739563,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.09220000356435776,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/T11273","display_name":"Advanced Graph Neural Networks","score":0.09139999747276306,"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":0.7080000042915344},{"id":"https://openalex.org/keywords/solver","display_name":"Solver","score":0.6978999972343445},{"id":"https://openalex.org/keywords/problem-solver","display_name":"Problem solver","score":0.5666000247001648},{"id":"https://openalex.org/keywords/limiting","display_name":"Limiting","score":0.5504000186920166},{"id":"https://openalex.org/keywords/quality","display_name":"Quality (philosophy)","score":0.4934000074863434},{"id":"https://openalex.org/keywords/code","display_name":"Code (set theory)","score":0.45590001344680786},{"id":"https://openalex.org/keywords/convergence","display_name":"Convergence (economics)","score":0.3774999976158142},{"id":"https://openalex.org/keywords/scheme","display_name":"Scheme (mathematics)","score":0.3612000048160553}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7289999723434448},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.7080000042915344},{"id":"https://openalex.org/C2778770139","wikidata":"https://www.wikidata.org/wiki/Q1966904","display_name":"Solver","level":2,"score":0.6978999972343445},{"id":"https://openalex.org/C3019612716","wikidata":"https://www.wikidata.org/wiki/Q730920","display_name":"Problem solver","level":2,"score":0.5666000247001648},{"id":"https://openalex.org/C188198153","wikidata":"https://www.wikidata.org/wiki/Q1613840","display_name":"Limiting","level":2,"score":0.5504000186920166},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.4934000074863434},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.48089998960494995},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.45590001344680786},{"id":"https://openalex.org/C2777303404","wikidata":"https://www.wikidata.org/wiki/Q759757","display_name":"Convergence (economics)","level":2,"score":0.3774999976158142},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3671000003814697},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.3612000048160553},{"id":"https://openalex.org/C164155591","wikidata":"https://www.wikidata.org/wiki/Q2067766","display_name":"Satisfiability modulo theories","level":2,"score":0.3343000113964081},{"id":"https://openalex.org/C139807058","wikidata":"https://www.wikidata.org/wiki/Q352374","display_name":"Adaptation (eye)","level":2,"score":0.3111000061035156},{"id":"https://openalex.org/C67203356","wikidata":"https://www.wikidata.org/wiki/Q1321905","display_name":"Reinforcement","level":2,"score":0.29760000109672546},{"id":"https://openalex.org/C107457646","wikidata":"https://www.wikidata.org/wiki/Q207434","display_name":"Human\u2013computer interaction","level":1,"score":0.29330000281333923},{"id":"https://openalex.org/C136197465","wikidata":"https://www.wikidata.org/wiki/Q1729295","display_name":"Variety (cybernetics)","level":2,"score":0.29190000891685486},{"id":"https://openalex.org/C77967617","wikidata":"https://www.wikidata.org/wiki/Q4677561","display_name":"Active learning (machine learning)","level":2,"score":0.2727000117301941},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.2685999870300293},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.26089999079704285},{"id":"https://openalex.org/C12713177","wikidata":"https://www.wikidata.org/wiki/Q1900281","display_name":"Perspective (graphical)","level":2,"score":0.260699987411499},{"id":"https://openalex.org/C2777212361","wikidata":"https://www.wikidata.org/wiki/Q5127848","display_name":"Class (philosophy)","level":2,"score":0.2581000030040741},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.2547999918460846}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.15224","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.15224","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.15224","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.15224","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"language":[1],"model-based":[2],"agents":[3],"make":[4],"mistakes,":[5],"yet":[6],"critique":[7,19],"can":[8],"often":[9],"guide":[10],"the":[11,22,28,37,56,102,112,133,148,156,212],"same":[12,29],"model":[13,23],"toward":[14],"correct":[15],"behavior.":[16,171],"However,":[17],"when":[18],"is":[20,98,216,228],"removed,":[21],"may":[24],"fail":[25],"again":[26],"on":[27,101,169,175,202,208],"query,":[30],"indicating":[31],"that":[32,76,126,155],"it":[33],"has":[34],"not":[35],"internalized":[36],"critique's":[38],"guidance":[39],"into":[40,92],"its":[41,50],"underlying":[42],"capability.":[43],"Meanwhile,":[44],"a":[45,73,79,82,85,122,139],"frozen":[46],"critic":[47,83,97,215],"cannot":[48],"improve":[49,160],"feedback":[51],"quality":[52],"over":[53,200],"time,":[54],"limiting":[55],"potential":[57],"for":[58],"iterative":[59],"self-improvement.":[60],"To":[61,110],"address":[62,111],"this,":[63],"we":[64],"propose":[65],"learning":[66],"to":[67,88,159,218],"internalize":[68],"self-critique":[69],"with":[70,132,194],"reinforcement":[71],"learning(ICRL),":[72],"novel":[74],"framework":[75],"jointly":[77],"trains":[78],"solver":[80,94,157],"and":[81,117,180,186,205],"from":[84],"shared":[86],"backbone":[87],"convert":[89],"critique-induced":[90],"success":[91],"unassisted":[93],"ability.":[95],"The":[96,226],"rewarded":[99],"based":[100],"solver's":[103,134],"subsequent":[104],"performance":[105],"gain,":[106],"incentivizing":[107],"actionable":[108],"feedback.":[109],"distribution":[113],"shift":[114],"between":[115],"critique-conditioned":[116,170],"critique-free":[118],"behavior,":[119],"ICRL":[120,174],"introduces":[121],"distribution-calibration":[123],"re-weighting":[124],"ratio":[125],"selectively":[127],"transfers":[128],"critique-guided":[129],"improvements":[130],"compatible":[131],"own":[135],"prompt":[136],"distribution.":[137],"Additionally,":[138],"role-wise":[140],"group":[141],"advantage":[142],"estimation":[143],"stabilizes":[144],"joint":[145],"optimization":[146],"across":[147],"two":[149],"roles.":[150],"Together,":[151],"these":[152],"mechanisms":[153],"ensure":[154],"learns":[158],"itself":[161],"without":[162],"external":[163],"critique,":[164],"rather":[165],"than":[166],"becoming":[167],"dependent":[168],"We":[172],"evaluate":[173],"diverse":[176],"benchmarks":[177],"spanning":[178],"agentic":[179,203],"mathematical":[181,209],"reasoning":[182],"tasks,":[183,204],"using":[184,222],"Qwen3-4B":[185],"Qwen3-8B":[187],"as":[188],"backbones.":[189],"Results":[190],"show":[191],"consistent":[192],"improvements,":[193],"average":[195],"gains":[196],"of":[197],"6.4":[198],"points":[199,207],"GRPO":[201],"7.0":[206],"reasoning.":[210],"Notably,":[211],"learned":[213],"8B":[214],"comparable":[217],"32B":[219],"critics":[220],"while":[221],"substantially":[223],"fewer":[224],"tokens.":[225],"code":[227],"available":[229],"at":[230],"https://github.com/brick-pid/ICRL.":[231]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-19T00:00:00"}