{"id":"https://openalex.org/W7161841810","doi":"https://doi.org/10.48550/arxiv.2605.19461","title":"Beyond Mode Collapse: Distribution Matching for Diverse Reasoning","display_name":"Beyond Mode Collapse: Distribution Matching for Diverse Reasoning","publication_year":2026,"publication_date":"2026-05-19","ids":{"openalex":"https://openalex.org/W7161841810","doi":"https://doi.org/10.48550/arxiv.2605.19461"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.19461","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.19461","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":"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.19461","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5136554547","display_name":"Xiaozhe Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Xiaozhe","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136534090","display_name":"Yang Li","orcid":"https://orcid.org/0000-0003-1682-0284"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Yang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136583151","display_name":"Xinyu Fang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Fang, Xinyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136533532","display_name":"Shengyuan Ding","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ding, Shengyuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136607630","display_name":"Peiji Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Peiji","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136506369","display_name":"Yongkang Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Yongkang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136515067","display_name":"Yichuan Ma","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ma, Yichuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136542580","display_name":"Tianyi Lyu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lyu, Tianyi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136556613","display_name":"Linyang Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Linyang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136575912","display_name":"Dahua Lin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lin, Dahua","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136529095","display_name":"Qipeng Guo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Guo, Qipeng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136582716","display_name":"Qingwen Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Qingwen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5136605771","display_name":"Kai Chen","orcid":"https://orcid.org/0000-0002-0164-1112"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Kai","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/T10462","display_name":"Reinforcement Learning in Robotics","score":0.8815000057220459,"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":0.8815000057220459,"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":0.017000000923871994,"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/T10653","display_name":"Robot Manipulation and Learning","score":0.014800000004470348,"subfield":{"id":"https://openalex.org/subfields/2207","display_name":"Control and Systems Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/matching","display_name":"Matching (statistics)","score":0.6463000178337097},{"id":"https://openalex.org/keywords/mode","display_name":"Mode (computer interface)","score":0.5777999758720398},{"id":"https://openalex.org/keywords/testbed","display_name":"Testbed","score":0.5406000018119812},{"id":"https://openalex.org/keywords/distribution","display_name":"Distribution (mathematics)","score":0.5062999725341797},{"id":"https://openalex.org/keywords/trajectory","display_name":"Trajectory","score":0.43220001459121704},{"id":"https://openalex.org/keywords/ideal","display_name":"Ideal (ethics)","score":0.42309999465942383},{"id":"https://openalex.org/keywords/quality","display_name":"Quality (philosophy)","score":0.41990000009536743},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.41780000925064087}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6473000049591064},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.6463000178337097},{"id":"https://openalex.org/C48677424","wikidata":"https://www.wikidata.org/wiki/Q6888088","display_name":"Mode (computer interface)","level":2,"score":0.5777999758720398},{"id":"https://openalex.org/C31395832","wikidata":"https://www.wikidata.org/wiki/Q1318674","display_name":"Testbed","level":2,"score":0.5406000018119812},{"id":"https://openalex.org/C110121322","wikidata":"https://www.wikidata.org/wiki/Q865811","display_name":"Distribution (mathematics)","level":2,"score":0.5062999725341797},{"id":"https://openalex.org/C13662910","wikidata":"https://www.wikidata.org/wiki/Q193139","display_name":"Trajectory","level":2,"score":0.43220001459121704},{"id":"https://openalex.org/C2776639384","wikidata":"https://www.wikidata.org/wiki/Q840396","display_name":"Ideal (ethics)","level":2,"score":0.42309999465942383},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.41990000009536743},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.41780000925064087},{"id":"https://openalex.org/C149441793","wikidata":"https://www.wikidata.org/wiki/Q200726","display_name":"Probability distribution","level":2,"score":0.41679999232292175},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.3991999924182892},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.3878999948501587},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.36800000071525574},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.35190001130104065},{"id":"https://openalex.org/C107673813","wikidata":"https://www.wikidata.org/wiki/Q812534","display_name":"Bayesian probability","level":2,"score":0.33410000801086426},{"id":"https://openalex.org/C52740198","wikidata":"https://www.wikidata.org/wiki/Q1539564","display_name":"Importance sampling","level":3,"score":0.3010999858379364},{"id":"https://openalex.org/C177769412","wikidata":"https://www.wikidata.org/wiki/Q278090","display_name":"Prior probability","level":3,"score":0.2924000024795532},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2865999937057495},{"id":"https://openalex.org/C167723999","wikidata":"https://www.wikidata.org/wiki/Q3773214","display_name":"Sampling distribution","level":2,"score":0.27720001339912415},{"id":"https://openalex.org/C115908005","wikidata":"https://www.wikidata.org/wiki/Q2668364","display_name":"Combinatorial explosion","level":2,"score":0.26910001039505005},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2614000141620636},{"id":"https://openalex.org/C77553402","wikidata":"https://www.wikidata.org/wiki/Q13222579","display_name":"Upper and lower bounds","level":2,"score":0.2540000081062317}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.19461","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.19461","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.19461","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.19461","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":"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"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"On-policy":[0],"reinforcement":[1],"learning":[2],"methods":[3],"like":[4],"GRPO":[5],"suffer":[6],"from":[7,34,102],"mode":[8,64,196],"collapse:":[9],"they":[10],"exhibit":[11],"reduced":[12],"solution":[13,21],"diversity,":[14],"concentrating":[15],"probability":[16],"mass":[17],"on":[18,116,143,151],"a":[19,50,75,128,190],"single":[20],"once":[22],"discovered":[23],"and":[24,149,158,170],"ceasing":[25],"exploration":[26,110,207],"of":[27,69],"alternative":[28],"strategies.":[29],"We":[30,56,113],"show":[31],"this":[32,93],"stems":[33],"reverse":[35],"KL":[36,71],"minimization's":[37],"mode-seeking":[38],"behavior,":[39],"which":[40,62],"reinforces":[41],"the":[42,89,103],"first":[43],"high-reward":[44],"trajectory":[45],"found":[46],"rather":[47],"than":[48],"maintaining":[49],"distribution":[51,79,91,187],"over":[52,80],"multiple":[53],"diverse":[54,209],"solutions.":[55],"propose":[57],"DMPO":[58,73,115,138],"(Distribution-Matching":[59],"Policy":[60],"Optimization),":[61],"prevents":[63],"collapse":[65,197],"through":[66],"principled":[67,192],"approximation":[68],"forward":[70],"minimization.":[72],"constructs":[74],"group":[76],"level":[77],"target":[78,106],"sampled":[81],"trajectories":[82],"proportional":[83],"to":[84,92,166,194],"their":[85],"rewards,":[86],"then":[87],"aligns":[88],"policy":[90],"target.":[94],"This":[95],"provides":[96],"mode-covering":[97],"behavior":[98],"without":[99],"requiring":[100],"sampling":[101],"intractable":[104],"global":[105],"distribution,":[107],"enabling":[108],"sustained":[109,206],"throughout":[111],"training.":[112],"validate":[114],"NP-hard":[117],"combinatorial":[118],"optimization,":[119],"where":[120],"exponentially":[121],"many":[122],"feasible":[123],"solutions":[124],"exist":[125],"but":[126],"only":[127],"few":[129],"approach":[130,193],"optimality,":[131],"an":[132],"ideal":[133],"testbed":[134],"for":[135],"evaluating":[136],"exploration.":[137],"achieves":[139],"43.9%":[140],"Quality":[141],"Ratio":[142],"text-based":[144],"NP-Bench":[145,153],"(vs.":[146,154],"GRPO's":[147],"40.1%)":[148],"43.1%":[150],"vision-based":[152],"38.4%),":[155],"demonstrating":[156,205],"9%":[157],"12%":[159],"relative":[160],"improvements":[161,204],"respectively.":[162],"These":[163],"gains":[164],"generalize":[165],"mathematical":[167],"reasoning":[168,180,210],"(+2.0%)":[169],"out-of-domain":[171],"tasks":[172],"(+2.3%),":[173],"showing":[174],"that":[175],"diversity-preserving":[176],"training":[177],"enhances":[178],"general":[179],"capabilities":[181],"across":[182,208],"modalities.":[183],"Our":[184],"work":[185],"establishes":[186],"matching":[188],"as":[189],"practical,":[191],"preventing":[195],"in":[198],"on-policy":[199],"RL,":[200],"with":[201],"consistent":[202],"quality":[203],"tasks.":[211]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-05-21T00:00:00"}