{"id":"https://openalex.org/W4416065894","doi":"https://doi.org/10.48550/arxiv.2509.07430","title":"The Choice of Divergence: A Neglected Key to Mitigating Diversity Collapse in Reinforcement Learning with Verifiable Reward","display_name":"The Choice of Divergence: A Neglected Key to Mitigating Diversity Collapse in Reinforcement Learning with Verifiable Reward","publication_year":2025,"publication_date":"2025-09-09","ids":{"openalex":"https://openalex.org/W4416065894","doi":"https://doi.org/10.48550/arxiv.2509.07430"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2509.07430","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2509.07430","pdf_url":"https://arxiv.org/pdf/2509.07430","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":null,"license_id":null,"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/2509.07430","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5100408749","display_name":"Li Long","orcid":"https://orcid.org/0000-0001-6036-2204"},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Li, Long","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5023615321","display_name":"Zhijian Zhou","orcid":"https://orcid.org/0000-0003-1340-1113"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhou, Zhijian","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5002910331","display_name":"Jiejie Hao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hao, Jiaran","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5061430474","display_name":"J Liu","orcid":"https://orcid.org/0009-0003-4286-0228"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Jason Klein","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5113373191","display_name":"Yanting Miao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Miao, Yanting","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5081644845","display_name":"Wei Pang","orcid":"https://orcid.org/0000-0002-1761-6659"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pang, Wei","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102869238","display_name":"Xiaoyu Tan","orcid":"https://orcid.org/0000-0003-3555-7143"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tan, Xiaoyu","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5034946345","display_name":"Wei-Ta Chu","orcid":"https://orcid.org/0000-0001-5722-7239"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chu, Wei","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100621319","display_name":"Zhe Wang","orcid":"https://orcid.org/0000-0002-3759-2041"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Zhe","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Pan, Shirui","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pan, Shirui","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5015442079","display_name":"Chao Qu","orcid":"https://orcid.org/0009-0003-4246-9106"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qu, Chao","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5107118224","display_name":"Yuan Qi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qi, Yuan","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":12,"corresponding_author_ids":["https://openalex.org/A5100408749"],"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/T10028","display_name":"Topic Modeling","score":0.21969999372959137,"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":0.21969999372959137,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.15760000050067902,"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":0.09700000286102295,"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/divergence","display_name":"Divergence (linguistics)","score":0.77920001745224},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.6970999836921692},{"id":"https://openalex.org/keywords/term","display_name":"Term (time)","score":0.6341999769210815},{"id":"https://openalex.org/keywords/verifiable-secret-sharing","display_name":"Verifiable secret sharing","score":0.6103000044822693},{"id":"https://openalex.org/keywords/function","display_name":"Function (biology)","score":0.5940999984741211},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.5458999872207642},{"id":"https://openalex.org/keywords/generator","display_name":"Generator (circuit theory)","score":0.5321000218391418},{"id":"https://openalex.org/keywords/diversity","display_name":"Diversity (politics)","score":0.39340001344680786}],"concepts":[{"id":"https://openalex.org/C207390915","wikidata":"https://www.wikidata.org/wiki/Q1230525","display_name":"Divergence (linguistics)","level":2,"score":0.77920001745224},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.6970999836921692},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6800000071525574},{"id":"https://openalex.org/C61797465","wikidata":"https://www.wikidata.org/wiki/Q1188986","display_name":"Term (time)","level":2,"score":0.6341999769210815},{"id":"https://openalex.org/C85847156","wikidata":"https://www.wikidata.org/wiki/Q59015987","display_name":"Verifiable secret sharing","level":3,"score":0.6103000044822693},{"id":"https://openalex.org/C14036430","wikidata":"https://www.wikidata.org/wiki/Q3736076","display_name":"Function (biology)","level":2,"score":0.5940999984741211},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.5458999872207642},{"id":"https://openalex.org/C2780992000","wikidata":"https://www.wikidata.org/wiki/Q17016113","display_name":"Generator (circuit theory)","level":3,"score":0.5321000218391418},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.39800000190734863},{"id":"https://openalex.org/C2781316041","wikidata":"https://www.wikidata.org/wiki/Q1230584","display_name":"Diversity (politics)","level":2,"score":0.39340001344680786},{"id":"https://openalex.org/C2780009758","wikidata":"https://www.wikidata.org/wiki/Q6804172","display_name":"Measure (data warehouse)","level":2,"score":0.37040001153945923},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.34380000829696655},{"id":"https://openalex.org/C2780771206","wikidata":"https://www.wikidata.org/wiki/Q3271761","display_name":"Safeguard","level":2,"score":0.34290000796318054},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.33889999985694885},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.3334999978542328},{"id":"https://openalex.org/C89611455","wikidata":"https://www.wikidata.org/wiki/Q6804646","display_name":"Mechanism (biology)","level":2,"score":0.3179999887943268},{"id":"https://openalex.org/C2778334786","wikidata":"https://www.wikidata.org/wiki/Q1586270","display_name":"Variation (astronomy)","level":2,"score":0.3140999972820282},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.31369999051094055},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.2727000117301941},{"id":"https://openalex.org/C2777212361","wikidata":"https://www.wikidata.org/wiki/Q5127848","display_name":"Class (philosophy)","level":2,"score":0.2522999942302704}],"mesh":[],"locations_count":3,"locations":[{"id":"pmh:oai:arXiv.org:2509.07430","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2509.07430","pdf_url":"https://arxiv.org/pdf/2509.07430","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"pmh:doi:10.48550/arxiv.2509.07430","is_oa":true,"landing_page_url":null,"pdf_url":null,"source":{"id":"https://openalex.org/S4406922384","display_name":"Open MIND","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"publisher-specific-oa","license_id":"https://openalex.org/licenses/publisher-specific-oa","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Article"},{"id":"doi:10.48550/arxiv.2509.07430","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2509.07430","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":"article"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2509.07430","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2509.07430","pdf_url":"https://arxiv.org/pdf/2509.07430","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4416065894.pdf","grobid_xml":"https://content.openalex.org/works/W4416065894.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"A":[0],"central":[1],"paradox":[2],"in":[3,26,124],"fine-tuning":[4],"Large":[5],"Language":[6],"Models":[7],"(LLMs)":[8],"with":[9,12],"Reinforcement":[10],"Learning":[11],"Verifiable":[13],"Reward":[14],"(RLVR)":[15],"is":[16,31,196,237],"the":[17,49,54,75,102,111,127,132,156,162,182,210,230],"frequent":[18],"degradation":[19,184],"of":[20,53,233],"multi-attempt":[21],"performance":[22],"(Pass@k)":[23],"despite":[24],"improvements":[25],"single-attempt":[27],"accuracy":[28],"(Pass@1).":[29],"This":[30],"often":[32],"accompanied":[33],"by":[34,100],"catastrophic":[35],"forgetting,":[36],"where":[37],"models":[38],"lose":[39],"previously":[40],"acquired":[41],"skills.":[42],"While":[43],"various":[44],"methods":[45],"have":[46,57],"been":[47,58],"proposed,":[48],"choice":[50],"and":[51,79,145,173,189,192,213,245],"function":[52,148],"divergence":[55,83,128,235],"term":[56,84,129],"surprisingly":[59],"unexamined":[60],"as":[61,131,149],"a":[62,82,88,121,150,221,234,238],"proactive":[63],"solution.":[64,133],"We":[65,119],"argue":[66],"that":[67,177,229],"standard":[68],"RLVR":[69],"objectives":[70],"--":[71,86],"both":[72,187],"those":[73,80],"using":[74,126,203],"mode-seeking":[76],"reverse":[77],"KL-divergence":[78],"forgoing":[81],"entirely":[85],"lack":[87],"crucial":[89],"mechanism":[90],"for":[91,225,241],"knowledge":[92,117],"retention.":[93],"The":[94],"reverse-KL":[95],"actively":[96],"accelerates":[97],"this":[98,159],"decay":[99],"narrowing":[101],"policy,":[103,158],"while":[104],"its":[105,115],"absence":[106],"provides":[107],"no":[108,214],"safeguard":[109],"against":[110],"model":[112,163],"drifting":[113],"from":[114,209],"diverse":[116,246],"base.":[118],"propose":[120],"fundamental":[122],"shift":[123],"perspective:":[125],"itself":[130],"Our":[134,218],"framework,":[135],"Diversity-Preserving":[136],"Hybrid":[137],"RL":[138],"(DPH-RL),":[139],"leverages":[140],"mass-covering":[141],"f-divergences":[142],"(like":[143],"forward-KL":[144],"JS-divergence)":[146],"to":[147,164],"rehearsal":[151],"mechanism.":[152],"By":[153],"continuously":[154],"referencing":[155],"initial":[157,211],"approach":[160],"forces":[161],"maintain":[165],"broad":[166],"solution":[167],"coverage.":[168],"Extensive":[169],"experiments":[170],"on":[171],"math":[172],"SQL":[174],"generation":[175],"demonstrate":[176],"DPH-RL":[178,195],"not":[179],"only":[180,207],"resolves":[181],"Pass@k":[183,190],"but":[185],"improves":[186],"Pass@1":[188],"in-":[191],"out-of-domain.":[193],"Additionally,":[194],"more":[197,243],"training-efficient":[198],"because":[199],"it":[200],"computes":[201],"f-divergence":[202],"generator":[204],"functions,":[205],"requiring":[206],"sampling":[208],"policy":[212],"online":[215],"reference":[216],"model.":[217],"work":[219],"highlights":[220],"crucial,":[222],"overlooked":[223],"axis":[224],"improving":[226],"RLVR,":[227],"demonstrating":[228],"proper":[231],"selection":[232],"measure":[236],"powerful":[239],"tool":[240],"building":[242],"general":[244],"reasoning":[247],"models.":[248]},"counts_by_year":[],"updated_date":"2026-04-16T08:26:57.006410","created_date":"2025-10-10T00:00:00"}