{"id":"https://openalex.org/W4415328612","doi":"https://doi.org/10.52202/085713-5096","title":"Pass@K Policy Optimization: Solving Harder Reinforcement Learning Problems","display_name":"Pass@K Policy Optimization: Solving Harder Reinforcement Learning Problems","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4415328612","doi":"https://doi.org/10.52202/085713-5096"},"language":"en","primary_location":{"id":"doi:10.52202/085713-5096","doi":"https://doi.org/10.52202/085713-5096","is_oa":false,"landing_page_url":"https://doi.org/10.52202/085713-5096","pdf_url":null,"source":{"id":"https://openalex.org/S4363606243","display_name":"neural information processing systems","issn_l":"1049-5258","issn":["1049-5258"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core","norway-2"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Advances in Neural Information Processing Systems 38","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2505.15201","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5063357214","display_name":"Christian Walder","orcid":"https://orcid.org/0000-0002-2411-3523","observed_orcids":["https://orcid.org/0000-0002-2411-3523"]},"institutions":[{"id":"https://openalex.org/I4210090411","display_name":"Google DeepMind (United Kingdom)","ror":"https://ror.org/00971b260","country_code":"GB","type":"company","lineage":["https://openalex.org/I4210090411","https://openalex.org/I4210128969"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Christian Walder","raw_affiliation_strings":["Google DeepMind"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google DeepMind","institution_ids":["https://openalex.org/I4210090411"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5045970638","display_name":"Deep Karkhanis","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4210090411","display_name":"Google DeepMind (United Kingdom)","ror":"https://ror.org/00971b260","country_code":"GB","type":"company","lineage":["https://openalex.org/I4210090411","https://openalex.org/I4210128969"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Deep Tejas Karkhanis","raw_affiliation_strings":["Google DeepMind"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google DeepMind","institution_ids":["https://openalex.org/I4210090411"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I4210090411"],"apc_list":null,"apc_paid":null,"fwci":0.9483,"has_fulltext":true,"cited_by_count":1,"citation_normalized_percentile":{"value":0.75585138,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":93,"max":96},"biblio":{"volume":null,"issue":null,"first_page":"169225","last_page":"169254"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10462","display_name":"Reinforcement Learning in Robotics","score":0.9990000128746033,"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":0.9990000128746033,"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/T12072","display_name":"Machine Learning and Algorithms","score":9.999999747378752e-05,"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/T12101","display_name":"Advanced Bandit Algorithms Research","score":9.999999747378752e-05,"subfield":{"id":"https://openalex.org/subfields/1803","display_name":"Management Science and Operations Research"},"field":{"id":"https://openalex.org/fields/18","display_name":"Decision Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/gradient-estimation","display_name":"gradient estimation","score":0.6549999713897705},{"id":"https://openalex.org/keywords/sample-diversity","display_name":"sample diversity","score":0.5049999952316284},{"id":"https://openalex.org/keywords/llm-fine-tuning","display_name":"LLM fine-tuning","score":0.640999972820282},{"id":"https://openalex.org/keywords/gemma-2","display_name":"Gemma 2","score":0.9120000004768372}],"concepts":[{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.7293999791145325},{"id":"https://openalex.org/C185429906","wikidata":"https://www.wikidata.org/wiki/Q1130160","display_name":"Estimator","level":2,"score":0.7253000140190125},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6607000231742859},{"id":"https://openalex.org/C126980161","wikidata":"https://www.wikidata.org/wiki/Q863783","display_name":"Simulated annealing","level":2,"score":0.5972999930381775},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.5820000171661377},{"id":"https://openalex.org/C196083921","wikidata":"https://www.wikidata.org/wiki/Q7915758","display_name":"Variance (accounting)","level":2,"score":0.5004000067710876},{"id":"https://openalex.org/C137836250","wikidata":"https://www.wikidata.org/wiki/Q984063","display_name":"Optimization problem","level":2,"score":0.48190000653266907},{"id":"https://openalex.org/C204241405","wikidata":"https://www.wikidata.org/wiki/Q461499","display_name":"Transformation (genetics)","level":3,"score":0.47360000014305115},{"id":"https://openalex.org/C48372109","wikidata":"https://www.wikidata.org/wiki/Q3913","display_name":"Binary number","level":2,"score":0.4000999927520752},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.3986000120639801},{"id":"https://openalex.org/C62644790","wikidata":"https://www.wikidata.org/wiki/Q3454689","display_name":"Variance reduction","level":3,"score":0.3653999865055084},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3560999929904938},{"id":"https://openalex.org/C188198153","wikidata":"https://www.wikidata.org/wiki/Q1613840","display_name":"Limiting","level":2,"score":0.35499998927116394},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.35260000824928284},{"id":"https://openalex.org/C198531522","wikidata":"https://www.wikidata.org/wiki/Q485146","display_name":"Sample (material)","level":2,"score":0.35179999470710754},{"id":"https://openalex.org/C52740198","wikidata":"https://www.wikidata.org/wiki/Q1539564","display_name":"Importance sampling","level":3,"score":0.3472999930381775},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.29980000853538513},{"id":"https://openalex.org/C2778915421","wikidata":"https://www.wikidata.org/wiki/Q3643177","display_name":"Performance improvement","level":2,"score":0.2624000012874603},{"id":"https://openalex.org/C2778334786","wikidata":"https://www.wikidata.org/wiki/Q1586270","display_name":"Variation (astronomy)","level":2,"score":0.25850000977516174},{"id":"https://openalex.org/C67203356","wikidata":"https://www.wikidata.org/wiki/Q1321905","display_name":"Reinforcement","level":2,"score":0.2556999921798706},{"id":"https://openalex.org/C41045048","wikidata":"https://www.wikidata.org/wiki/Q202843","display_name":"Linear programming","level":2,"score":0.25369998812675476},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.25029999017715454}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.52202/085713-5096","doi":"https://doi.org/10.52202/085713-5096","is_oa":false,"landing_page_url":"https://doi.org/10.52202/085713-5096","pdf_url":null,"source":{"id":"https://openalex.org/S4363606243","display_name":"neural information processing systems","issn_l":"1049-5258","issn":["1049-5258"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core","norway-2"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Advances in Neural Information Processing Systems 38","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2505.15201","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2505.15201","pdf_url":"https://arxiv.org/pdf/2505.15201","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2505.15201","doi":"https://doi.org/10.48550/arxiv.2505.15201","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2505.15201","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,"listed_in":[],"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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2505.15201","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2505.15201","pdf_url":"https://arxiv.org/pdf/2505.15201","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4415328612.pdf","grobid_xml":"https://content.openalex.org/works/W4415328612.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W6948538263","https://openalex.org/W4416550392","https://openalex.org/W4417098490","https://openalex.org/W7101911636","https://openalex.org/W7220933582","https://openalex.org/W7170567086","https://openalex.org/W7165921964","https://openalex.org/W7131407993","https://openalex.org/W7101782290","https://openalex.org/W4415238102"],"abstract_inverted_index":{"Reinforcement":[0],"Learning":[1],"(RL)":[2],"algorithms":[3],"sample":[4],"multiple":[5],"n>1":[6],"solution":[7],"attempts":[8],"for":[9,18,79,99,153,166,223,248],"each":[10],"problem":[11],"and":[12,21,34,48,101,108,132,179,234,244],"reward":[13,85,110,192],"them":[14],"independently.":[15],"This":[16,41],"optimizes":[17,222],"pass@1":[19,164,183,243,254],"performance":[20,165],"prioritizes":[22],"the":[23,29,32,43,66,106,145,199,212,224,242,272],"strength":[24],"of":[25,31,37,39,74,81,151,161,203,274],"isolated":[26],"samples":[27,82],"at":[28],"expense":[30],"diversity":[33],"collective":[35],"utility":[36,270,273],"sets":[38,80,251],"samples.":[40,276],"under-utilizes":[42],"sampling":[44],"capacity,":[45],"limiting":[46],"exploration":[47,266],"eventual":[49],"improvement":[50],"on":[51,65,194],"harder":[52,235],"examples.":[53],"As":[54],"a":[55,63,130],"fix,":[56],"we":[57],"propose":[58],"Pass-at-k":[59],"Policy":[60],"Optimization":[61],"(PKPO),":[62],"transformation":[64,134,220],"final":[67],"rewards":[68,123],"which":[69,197],"leads":[70],"to":[71,92,119,141,147,264],"direct":[72],"optimization":[73,114,150,255],"pass@k":[75,100,152,167,187,245,258],"performance,":[76],"thus":[77],"optimizing":[78,176],"that":[83,124,218],"maximize":[84],"when":[86],"considered":[87],"jointly.":[88],"Our":[89],"contribution":[90],"is":[91,144],"derive":[93],"novel":[94],"low":[95],"variance":[96,200],"unbiased":[97],"estimators":[98,117],"its":[102],"gradient,":[103],"in":[104],"both":[105,177,241],"binary":[107],"continuous":[109],"settings.":[111],"We":[112,189,206,216],"show":[113],"with":[115,122],"our":[116,169,191,204,219,257],"reduces":[118],"standard":[120],"RL":[121],"have":[125],"been":[126],"jointly":[127],"transformed":[128],"by":[129,267],"stable":[131],"efficient":[133],"function.":[135],"While":[136],"previous":[137],"efforts":[138],"are":[139],"restricted":[140],"k=n,":[142],"ours":[143],"first":[146],"enable":[148,231],"robust":[149],"any":[154],"arbitrary":[155],"k":[156,173,229,239],"<=":[157],"n.":[158],"Moreover,":[159],"instead":[160],"trading":[162],"off":[163],"gains,":[168],"method":[170],"allows":[171],"annealing":[172,238],"during":[174],"training,":[175],"metrics":[178],"often":[180],"achieving":[181],"strong":[182],"numbers":[184],"alongside":[185],"significant":[186],"gains.":[188],"validate":[190],"transformations":[193],"toy":[195],"experiments,":[196],"reveal":[198],"reducing":[201],"properties":[202],"formulations.":[205],"also":[207],"include":[208],"real-world":[209],"examples":[210],"using":[211],"open-source":[213],"LLM,":[214],"GEMMA-2.":[215],"find":[217],"effectively":[221],"target":[225],"k.":[226],"Furthermore,":[227],"higher":[228],"values":[230],"solving":[232],"more":[233],"problems,":[236],"while":[237],"boosts":[240],".":[246],"Crucially,":[247],"challenging":[249],"task":[250],"where":[252],"conventional":[253],"stalls,":[256],"approach":[259],"unblocks":[260],"learning,":[261],"likely":[262],"due":[263],"better":[265],"prioritizing":[268],"joint":[269],"over":[271],"individual":[275]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-10-09T09:14:12.587421","created_date":"2025-10-19T00:00:00"}