{"id":"https://openalex.org/W7166875078","doi":"https://doi.org/10.18653/v1/2026.findings-acl.467","title":"MMR-GRPO: Accelerating GRPO-Style Training through Diversity-Aware Reward Reweighting","display_name":"MMR-GRPO: Accelerating GRPO-Style Training through Diversity-Aware Reward Reweighting","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7166875078","doi":"https://doi.org/10.18653/v1/2026.findings-acl.467"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.findings-acl.467","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.467","pdf_url":"https://aclanthology.org/2026.findings-acl.467.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":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.findings-acl.467.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5089811959","display_name":"Kangda Wei","orcid":null},"institutions":[{"id":"https://openalex.org/I91045830","display_name":"Texas A&M University","ror":"https://ror.org/01f5ytq51","country_code":"US","type":"education","lineage":["https://openalex.org/I91045830"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Kangda Wei","raw_affiliation_strings":["Department of Computer Science and Engineering Texas A&M University , College Station , TX"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science and Engineering Texas A&M University , College Station , TX","institution_ids":["https://openalex.org/I91045830"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5101688218","display_name":"Ruihong Huang","orcid":"https://orcid.org/0000-0002-4639-7766"},"institutions":[{"id":"https://openalex.org/I91045830","display_name":"Texas A&M University","ror":"https://ror.org/01f5ytq51","country_code":"US","type":"education","lineage":["https://openalex.org/I91045830"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Ruihong Huang","raw_affiliation_strings":["Department of Computer Science and Engineering Texas A&M University , College Station , TX"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science and Engineering Texas A&M University , College Station , TX","institution_ids":["https://openalex.org/I91045830"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I91045830"],"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":"9584","last_page":"9605"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12535","display_name":"Machine Learning and Data Classification","score":0.09619999676942825,"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/T12535","display_name":"Machine Learning and Data Classification","score":0.09619999676942825,"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/T10667","display_name":"Emotion and Mood Recognition","score":0.059700001031160355,"subfield":{"id":"https://openalex.org/subfields/3205","display_name":"Experimental and Cognitive Psychology"},"field":{"id":"https://openalex.org/fields/32","display_name":"Psychology"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.04430000111460686,"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/training","display_name":"Training (meteorology)","score":0.5722000002861023},{"id":"https://openalex.org/keywords/noise","display_name":"Noise (video)","score":0.27140000462532043},{"id":"https://openalex.org/keywords/field","display_name":"Field (mathematics)","score":0.24320000410079956},{"id":"https://openalex.org/keywords/training-set","display_name":"Training set","score":0.2362000048160553},{"id":"https://openalex.org/keywords/feature","display_name":"Feature (linguistics)","score":0.23280000686645508}],"concepts":[{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.5722000002861023},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5184999704360962},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4284999966621399},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.27140000462532043},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2685999870300293},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.249099999666214},{"id":"https://openalex.org/C9652623","wikidata":"https://www.wikidata.org/wiki/Q190109","display_name":"Field (mathematics)","level":2,"score":0.24320000410079956},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.2362000048160553},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.23280000686645508},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.2280000001192093}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.findings-acl.467","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.467","pdf_url":"https://aclanthology.org/2026.findings-acl.467.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"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.findings-acl.467","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.467","pdf_url":"https://aclanthology.org/2026.findings-acl.467.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":[{"id":"https://openalex.org/G3004366879","display_name":"CAREER: Discourse Level Event-Event Relation Identification","funder_award_id":"1942918","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":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7166875078.pdf","grobid_xml":"https://content.openalex.org/works/W7166875078.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Group":[0],"Relative":[1],"Policy":[2],"Optimization":[3],"(GRPO)":[4],"(Shao":[5],"et":[6,46],"al.,":[7,47],"2024)":[8],"has":[9,33],"become":[10],"a":[11,91],"standard":[12],"approach":[13],"for":[14],"training":[15,28,38,52,137],"mathematical":[16,121],"reasoning":[17,122],"models;":[18],"however,":[19],"its":[20],"reliance":[21],"on":[22,76,133],"multiple":[23],"completions":[24],"per":[25],"prompt":[26],"makes":[27],"computationally":[29],"expensive.Although":[30],"recent":[31],"work":[32],"reduced":[34],"the":[35,49],"number":[36],"of":[37],"steps":[39,138],"required":[40],"to":[41,61,72,89],"reach":[42],"peak":[43,129],"performance":[44,130],"(Yu":[45],"2025),":[48],"overall":[50],"wall-clock":[51,142],"time":[53],"often":[54],"remains":[55],"unchanged":[56],"or":[57,83],"even":[58],"increases":[59],"due":[60],"higher":[62],"per-step":[63],"cost.We":[64],"propose":[65],"MMR-GRPO,":[66],"which":[67],"integrates":[68],"Maximal":[69],"Marginal":[70],"Relevance":[71],"reweigh":[73],"rewards":[74],"based":[75],"completion":[77],"diversity.The":[78],"rationale":[79],"is":[80,153],"that":[81,125],"redundant":[82],"similar":[84],"completions,":[85],"if":[86],"repeatedly":[87],"used":[88],"train":[90],"model,":[92],"will":[93],"create":[94],"an":[95],"\"exploitation":[96],"trap\"":[97],"and":[98,119,139,150],"slow":[99],"down":[100],"model":[101,111],"convergence":[102],"in":[103],"GRPO":[104,117],"style":[105],"reinforcement":[106],"learning.Extensive":[107],"evaluations":[108],"across":[109,147],"three":[110,116],"sizes":[112],"(1.5B,":[113],"7B,":[114],"8B),":[115],"variants,":[118],"five":[120],"benchmarks":[123],"show":[124],"MMR-GRPO":[126],"achieves":[127],"comparable":[128],"while":[131],"requiring":[132],"average":[134],"47.9%":[135],"fewer":[136],"70.2%":[140],"less":[141],"time.These":[143],"gains":[144],"are":[145],"consistent":[146],"models,":[148],"methods,":[149],"benchmarks.Our":[151],"code":[152],"released":[154],"at:":[155],"https:":[156]},"counts_by_year":[],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2026-07-02T00:00:00"}