{"id":"https://openalex.org/W7163149836","doi":"https://doi.org/10.48550/arxiv.2606.00544","title":"Escaping the Mode Lottery: Multi-Response Training Improves Language Model Generalization","display_name":"Escaping the Mode Lottery: Multi-Response Training Improves Language Model Generalization","publication_year":2026,"publication_date":"2026-05-30","ids":{"openalex":"https://openalex.org/W7163149836","doi":"https://doi.org/10.48550/arxiv.2606.00544"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.00544","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.00544","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":"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.2606.00544","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5134156798","display_name":"Hasan Amin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Amin, Hasan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5076242590","display_name":"Kian Ahrabian","orcid":"https://orcid.org/0000-0002-6606-9735"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ahrabian, Kian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137690460","display_name":"Ming Yin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yin, Ming","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5124335891","display_name":"Rajiv Khanna","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Khanna, Rajiv","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":0.3621000051498413,"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":0.3621000051498413,"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":0.2992999851703644,"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/T11636","display_name":"Artificial Intelligence in Healthcare and Education","score":0.033799998462200165,"subfield":{"id":"https://openalex.org/subfields/2718","display_name":"Health Informatics"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/submodular-set-function","display_name":"Submodular set function","score":0.618399977684021},{"id":"https://openalex.org/keywords/generalization","display_name":"Generalization","score":0.6074000000953674},{"id":"https://openalex.org/keywords/mode","display_name":"Mode (computer interface)","score":0.5787000060081482},{"id":"https://openalex.org/keywords/selection","display_name":"Selection (genetic algorithm)","score":0.5774999856948853},{"id":"https://openalex.org/keywords/variance","display_name":"Variance (accounting)","score":0.5533000230789185},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.42250001430511475},{"id":"https://openalex.org/keywords/statistical-model","display_name":"Statistical model","score":0.396699994802475},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.37299999594688416}],"concepts":[{"id":"https://openalex.org/C178621042","wikidata":"https://www.wikidata.org/wiki/Q7631710","display_name":"Submodular set function","level":2,"score":0.618399977684021},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6080999970436096},{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.6074000000953674},{"id":"https://openalex.org/C48677424","wikidata":"https://www.wikidata.org/wiki/Q6888088","display_name":"Mode (computer interface)","level":2,"score":0.5787000060081482},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.5774999856948853},{"id":"https://openalex.org/C196083921","wikidata":"https://www.wikidata.org/wiki/Q7915758","display_name":"Variance (accounting)","level":2,"score":0.5533000230789185},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.42250001430511475},{"id":"https://openalex.org/C114289077","wikidata":"https://www.wikidata.org/wiki/Q3284399","display_name":"Statistical model","level":2,"score":0.396699994802475},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.37299999594688416},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.3702999949455261},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.36070001125335693},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.3398999869823456},{"id":"https://openalex.org/C93959086","wikidata":"https://www.wikidata.org/wiki/Q6888345","display_name":"Model selection","level":2,"score":0.33809998631477356},{"id":"https://openalex.org/C43214815","wikidata":"https://www.wikidata.org/wiki/Q7310987","display_name":"Reliability (semiconductor)","level":3,"score":0.31349998712539673},{"id":"https://openalex.org/C149782125","wikidata":"https://www.wikidata.org/wiki/Q160039","display_name":"Econometrics","level":1,"score":0.31290000677108765},{"id":"https://openalex.org/C50335755","wikidata":"https://www.wikidata.org/wiki/Q483247","display_name":"Phenomenon","level":2,"score":0.3018999993801117},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3003000020980835},{"id":"https://openalex.org/C43555835","wikidata":"https://www.wikidata.org/wiki/Q2300258","display_name":"Conditional probability distribution","level":2,"score":0.29109999537467957},{"id":"https://openalex.org/C149441793","wikidata":"https://www.wikidata.org/wiki/Q200726","display_name":"Probability distribution","level":2,"score":0.2896000146865845},{"id":"https://openalex.org/C177769412","wikidata":"https://www.wikidata.org/wiki/Q278090","display_name":"Prior probability","level":3,"score":0.2720000147819519},{"id":"https://openalex.org/C149364088","wikidata":"https://www.wikidata.org/wiki/Q185917","display_name":"Translation (biology)","level":4,"score":0.26910001039505005}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.00544","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.00544","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":"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.2606.00544","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.00544","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":"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":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Modern":[0],"language-model":[1],"fine-tuning":[2],"typically":[3],"pairs":[4],"each":[5],"prompt":[6],"with":[7,164,187,207,225],"a":[8,22,27,30,40,62,103,156,177,197,222,240,243],"single":[9],"response,":[10],"even":[11],"though":[12],"many":[13],"prompts":[14,76,84],"admit":[15],"multiple":[16,56,110],"valid":[17],"completions.":[18],"This":[19,101],"effectively":[20],"reduces":[21],"multi-modal":[23],"conditional":[24,98],"distribution":[25],"to":[26],"one-sample":[28],"view,":[29],"phenomenon":[31],"we":[32],"call":[33],"the":[34,88,97,143,170,188,208],"\"mode":[35],"lottery,\"":[36],"where":[37,181],"training":[38,52],"emphasizes":[39],"subset":[41],"of":[42,65],"plausible":[43],"modes":[44],"while":[45,91],"leaving":[46],"others":[47],"underrepresented.":[48],"We":[49,133],"study":[50],"multi-response":[51,131,200],"(MRT),":[53],"which":[54],"retains":[55],"responses":[57,78,93,111,229],"per":[58],"prompt,":[59],"and":[60,67,77,121,138,155,173,193,232],"develop":[61],"principled":[63],"account":[64],"when":[66,108,228],"why":[68,123],"it":[69],"helps.":[70],"Our":[71],"key":[72],"insight":[73],"is":[74,112,142,238],"that":[75,106,140],"are":[79,230],"distinct":[80],"statistical":[81],"resources:":[82],"additional":[83,92],"reduce":[85,94],"uncertainty":[86,95,119],"about":[87,96],"input":[89],"distribution,":[90],"output":[99],"distribution.":[100],"yields":[102],"variance-budget":[104],"tradeoff":[105],"predicts":[107],"retaining":[109],"worthwhile,":[113],"shows":[114],"diminishing":[115],"returns":[116],"as":[117,221],"prompt-level":[118],"dominates,":[120],"explains":[122],"large":[124],"redundant":[125],"corpora":[126],"can":[127,151],"exhibit":[128],"an":[129,161],"implicit":[130],"effect.":[132],"further":[134],"analyze":[135],"response":[136,219],"selection,":[137],"show":[139],"Random-K-of-N":[141],"unbiased":[144],"default":[145],"for":[146],"distributional":[147,205],"fine-tuning,":[148],"reward-based":[149],"selection":[150,174,183],"induce":[152],"mode":[153,180],"collapse,":[154],"submodular":[157],"quality-diversity":[158],"objective":[159],"provides":[160],"efficient":[162],"alternative":[163],"theoretical":[165],"guarantees.":[166],"Controlled":[167],"simulations":[168],"validate":[169],"predicted":[171],"variance":[172],"effects,":[175],"including":[176,196],"striking":[178],"failure":[179],"reward-only":[182],"produces":[184],"gradients":[185],"misaligned":[186],"true":[189],"objective.":[190],"Across":[191],"structured":[192],"real-world":[194],"datasets,":[195],"new":[198],"multi-prompt,":[199],"benchmark,":[201],"MRT":[202,217],"consistently":[203],"improves":[204],"generalization,":[206],"largest":[209],"gains":[210],"in":[211],"high":[212],"response-diversity,":[213],"low":[214],"prompt-redundancy":[215],"regimes.":[216],"reframes":[218],"multiplicity":[220],"data-allocation":[223],"problem":[224],"clear":[226],"guidance:":[227],"cheap":[231],"diverse,":[233],"keeping":[234],"more":[235],"than":[236],"one":[237],"not":[239],"heuristic,":[241],"but":[242],"statistically":[244],"grounded":[245],"choice.":[246]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-03T00:00:00"}