{"id":"https://openalex.org/W4416681949","doi":"https://doi.org/10.18653/v1/2026.findings-acl.1291","title":"Learning to Refine: Self-Refinement of Parallel Reasoning in LLMs","display_name":"Learning to Refine: Self-Refinement of Parallel Reasoning in LLMs","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W4416681949","doi":"https://doi.org/10.18653/v1/2026.findings-acl.1291"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2026.findings-acl.1291","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.1291","pdf_url":"https://aclanthology.org/2026.findings-acl.1291.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.findings-acl.1291.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5069573368","display_name":"Qibin Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qibin Wang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5120665832","display_name":"Pu Zhao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pu Zhao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5061624006","display_name":"Shaohan Huang","orcid":"https://orcid.org/0000-0003-4324-6337"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shaohan Huang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101564849","display_name":"Fangkai Yang","orcid":"https://orcid.org/0000-0002-3089-0345"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Fangkai Yang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100364381","display_name":"Lu Wang","orcid":"https://orcid.org/0000-0001-6345-3873"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lu Wang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5120519568","display_name":"Furu Wei","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Furu Wei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5088646345","display_name":"Qingwei Lin","orcid":"https://orcid.org/0000-0003-2559-2383"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qingwei Lin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5070722259","display_name":"Saravan Rajmohan","orcid":"https://orcid.org/0000-0002-2019-213X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Saravan Rajmohan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5101567180","display_name":"Dongmei Zhang","orcid":"https://orcid.org/0000-0001-6302-0373"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dongmei Zhang","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":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.02854804,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"25904","last_page":"25921"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.6335999965667725,"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":0.6335999965667725,"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/T11273","display_name":"Advanced Graph Neural Networks","score":0.0771000012755394,"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":0.04569999873638153,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/set","display_name":"Set (abstract data type)","score":0.5889000296592712},{"id":"https://openalex.org/keywords/pipeline","display_name":"Pipeline (software)","score":0.5493000149726868},{"id":"https://openalex.org/keywords/voting","display_name":"Voting","score":0.44040000438690186},{"id":"https://openalex.org/keywords/scaling","display_name":"Scaling","score":0.4271000027656555},{"id":"https://openalex.org/keywords/quality","display_name":"Quality (philosophy)","score":0.39469999074935913},{"id":"https://openalex.org/keywords/generative-model","display_name":"Generative model","score":0.3628000020980835},{"id":"https://openalex.org/keywords/generative-grammar","display_name":"Generative grammar","score":0.36070001125335693}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7200999855995178},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.5889000296592712},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.5493000149726868},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5325999855995178},{"id":"https://openalex.org/C520049643","wikidata":"https://www.wikidata.org/wiki/Q189760","display_name":"Voting","level":3,"score":0.44040000438690186},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.4271000027656555},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4271000027656555},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.39469999074935913},{"id":"https://openalex.org/C167966045","wikidata":"https://www.wikidata.org/wiki/Q5532625","display_name":"Generative model","level":3,"score":0.3628000020980835},{"id":"https://openalex.org/C39890363","wikidata":"https://www.wikidata.org/wiki/Q36108","display_name":"Generative grammar","level":2,"score":0.36070001125335693},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.31769999861717224},{"id":"https://openalex.org/C105339364","wikidata":"https://www.wikidata.org/wiki/Q2297740","display_name":"Software deployment","level":2,"score":0.29440000653266907},{"id":"https://openalex.org/C104114177","wikidata":"https://www.wikidata.org/wiki/Q79782","display_name":"Motion (physics)","level":2,"score":0.2865000069141388},{"id":"https://openalex.org/C45493050","wikidata":"https://www.wikidata.org/wiki/Q7884934","display_name":"Unified Model","level":2,"score":0.27239999175071716},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.27059999108314514},{"id":"https://openalex.org/C20162079","wikidata":"https://www.wikidata.org/wiki/Q1151406","display_name":"Case-based reasoning","level":2,"score":0.2705000042915344},{"id":"https://openalex.org/C153668964","wikidata":"https://www.wikidata.org/wiki/Q27636","display_name":"Majority rule","level":2,"score":0.2587999999523163}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.18653/v1/2026.findings-acl.1291","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.1291","pdf_url":"https://aclanthology.org/2026.findings-acl.1291.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2509.00084","is_oa":true,"landing_page_url":"https://arxiv.org/abs/2509.00084","pdf_url":"https://arxiv.org/pdf/2509.00084","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,"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2509.00084","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2509.00084","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,"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:10.18653/v1/2026.findings-acl.1291","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.1291","pdf_url":"https://aclanthology.org/2026.findings-acl.1291.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4416681949.pdf","grobid_xml":"https://content.openalex.org/works/W4416681949.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Test-time":[0],"scaling":[1,100,125],"(TTS)":[2],"has":[3],"gained":[4],"widespread":[5],"attention":[6],"for":[7],"enhancing":[8],"LLM":[9],"reasoning.Existing":[10],"approaches":[11],"such":[12],"as":[13,20],"Best-of-N":[14],"and":[15,47,105,153,190,192],"majority":[16,90],"voting":[17],"are":[18,41],"limited":[19],"their":[21],"performance":[22,170],"depends":[23],"on":[24,53,114,160],"the":[25,60,78,84,94,111,129,181],"quality":[26],"of":[27,87],"candidate":[28],"responses,":[29],"making":[30],"them":[31],"unable":[32],"to":[33,82,149,196],"produce":[34],"a":[35,49,56,75,98,122,146,155],"correct":[36],"solution":[37],"when":[38],"all":[39],"candidates":[40,46,152],"incorrect.Parallel":[42],"self-refinement,":[43],"generating":[44],"multiple":[45,187],"synthesizing":[48],"refined":[50],"answer":[51,158],"conditioned":[52],"them,":[54],"offers":[55],"promising":[57],"alternative,":[58],"but":[59],"underlying":[61],"mechanism":[62],"driving":[63],"its":[64],"effectiveness":[65],"remains":[66],"obscure.To":[67],"bridge":[68],"this":[69,115],"gap":[70,139],"in":[71],"understanding,":[72],"we":[73,117],"introduce":[74],"new":[76],"metric,":[77],"Refinement":[79,95],"Gap,":[80],"designed":[81],"quantify":[83],"relative":[85],"improvement":[86],"self-refinement":[88],"beyond":[89],"voting.We":[91],"show":[92],"that":[93,127,165],"Gap":[96],"exhibits":[97,193],"clear":[99],"trend":[101],"with":[102,110,136],"model":[103,148,188],"size":[104],"is":[106],"only":[107],"weakly":[108],"correlated":[109],"base":[112],"capability.Based":[113],"discovery,":[116],"propose":[118],"Generative":[119],"Self-Refinement":[120],"(GSR),":[121],"parallel":[123,177],"test-time":[124],"framework":[126],"transfers":[128,185],"refinement":[130,138,183],"policy":[131],"from":[132],"larger":[133],"teacher":[134],"models":[135],"higher":[137],"into":[140],"smaller":[141],"students.Crucially,":[142],"GSR":[143],"jointly":[144],"trains":[145],"single":[147],"generate":[150],"strong":[151],"refine":[154],"better":[156],"final":[157],"based":[159],"these":[161],"candidates.Experimental":[162],"results":[163],"demonstrate":[164],"our":[166],"method":[167],"achieves":[168],"state-of-the-art":[169],"across":[171,186],"five":[172],"mathematical":[173],"benchmarks":[174],"over":[175],"other":[176],"aggregation":[178],"methods,":[179],"while":[180],"learned":[182],"skill":[184],"scales":[189],"families":[191],"robust":[194],"generalization":[195],"an":[197],"out-of-distribution":[198],"domain.":[199]},"counts_by_year":[],"updated_date":"2026-08-05T07:39:15.569665","created_date":"2025-10-10T00:00:00"}