{"id":"https://openalex.org/W7125773359","doi":"https://doi.org/10.48550/arxiv.2601.18795","title":"Reuse your FLOPs: Scaling RL on Hard Problems by Conditioning on Very Off-Policy Prefixes","display_name":"Reuse your FLOPs: Scaling RL on Hard Problems by Conditioning on Very Off-Policy Prefixes","publication_year":2026,"publication_date":"2026-01-26","ids":{"openalex":"https://openalex.org/W7125773359","doi":"https://doi.org/10.48550/arxiv.2601.18795"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2601.18795","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2601.18795","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.2601.18795","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5123892397","display_name":"Amrith Setlur","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Setlur, Amrith","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123886356","display_name":"Zijian Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Zijian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123909293","display_name":"Andrew Cohen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cohen, Andrew","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123965830","display_name":"Paria Rashidinejad","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Rashidinejad, Paria","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5069097726","display_name":"Sang Michael Xie","orcid":"https://orcid.org/0000-0002-0820-2753"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xie, Sang Michael","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":0.5710999965667725,"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":0.5710999965667725,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.125900000333786,"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/T10028","display_name":"Topic Modeling","score":0.034299999475479126,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/reuse","display_name":"Reuse","score":0.7677000164985657},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.6940000057220459},{"id":"https://openalex.org/keywords/flexibility","display_name":"Flexibility (engineering)","score":0.6876999735832214},{"id":"https://openalex.org/keywords/prefix","display_name":"Prefix","score":0.569100022315979},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5616000294685364},{"id":"https://openalex.org/keywords/sampling","display_name":"Sampling (signal processing)","score":0.49070000648498535},{"id":"https://openalex.org/keywords/conditioning","display_name":"Conditioning","score":0.4740000069141388},{"id":"https://openalex.org/keywords/sample","display_name":"Sample (material)","score":0.4593999981880188}],"concepts":[{"id":"https://openalex.org/C206588197","wikidata":"https://www.wikidata.org/wiki/Q846574","display_name":"Reuse","level":2,"score":0.7677000164985657},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.6940000057220459},{"id":"https://openalex.org/C2780598303","wikidata":"https://www.wikidata.org/wiki/Q65921492","display_name":"Flexibility (engineering)","level":2,"score":0.6876999735832214},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6865000128746033},{"id":"https://openalex.org/C141603448","wikidata":"https://www.wikidata.org/wiki/Q134830","display_name":"Prefix","level":2,"score":0.569100022315979},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5616000294685364},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.49070000648498535},{"id":"https://openalex.org/C45262634","wikidata":"https://www.wikidata.org/wiki/Q5159291","display_name":"Conditioning","level":2,"score":0.4740000069141388},{"id":"https://openalex.org/C198531522","wikidata":"https://www.wikidata.org/wiki/Q485146","display_name":"Sample (material)","level":2,"score":0.4593999981880188},{"id":"https://openalex.org/C42058472","wikidata":"https://www.wikidata.org/wiki/Q810214","display_name":"Base (topology)","level":2,"score":0.41499999165534973},{"id":"https://openalex.org/C106195933","wikidata":"https://www.wikidata.org/wiki/Q7847935","display_name":"Truncation (statistics)","level":2,"score":0.4108999967575073},{"id":"https://openalex.org/C73602740","wikidata":"https://www.wikidata.org/wiki/Q7795822","display_name":"Thompson sampling","level":3,"score":0.4083000123500824},{"id":"https://openalex.org/C2776175482","wikidata":"https://www.wikidata.org/wiki/Q1195816","display_name":"Transfer (computing)","level":2,"score":0.40119999647140503},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.39969998598098755},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3959999978542328},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.38510000705718994},{"id":"https://openalex.org/C150899416","wikidata":"https://www.wikidata.org/wiki/Q1820378","display_name":"Transfer of learning","level":2,"score":0.3653999865055084},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.34950000047683716},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.32109999656677246},{"id":"https://openalex.org/C2779843651","wikidata":"https://www.wikidata.org/wiki/Q7390335","display_name":"SIGNAL (programming language)","level":2,"score":0.30790001153945923},{"id":"https://openalex.org/C52740198","wikidata":"https://www.wikidata.org/wiki/Q1539564","display_name":"Importance sampling","level":3,"score":0.27869999408721924},{"id":"https://openalex.org/C12725497","wikidata":"https://www.wikidata.org/wiki/Q810247","display_name":"Baseline (sea)","level":2,"score":0.275299996137619},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.25940001010894775},{"id":"https://openalex.org/C94375191","wikidata":"https://www.wikidata.org/wiki/Q11205","display_name":"Arithmetic","level":1,"score":0.25850000977516174}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2601.18795","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2601.18795","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.2601.18795","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2601.18795","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":[{"id":"https://metadata.un.org/sdg/16","score":0.5343291163444519,"display_name":"Peace, Justice and strong institutions"}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Typical":[0],"reinforcement":[1],"learning":[2,23,86],"(RL)":[3],"methods":[4,50],"for":[5,191],"LLM":[6],"reasoning":[7,168],"waste":[8],"compute":[9,193],"on":[10,66,88,129,183,195],"hard":[11,89,167],"problems,":[12,169],"where":[13,63],"correct":[14],"on-policy":[15,75],"traces":[16,72,154,220],"are":[17,221],"rare,":[18],"policy":[19],"gradients":[20],"vanish,":[21],"and":[22,73,200,213],"stalls.":[24],"To":[25],"bootstrap":[26],"more":[27,120],"efficient":[28],"RL,":[29],"we":[30,64,124,150],"consider":[31],"reusing":[32],"old":[33],"sampling":[34,157],"FLOPs":[35],"(from":[36],"prior":[37],"inference":[38],"or":[39],"RL":[40,58,76,116],"training)":[41],"in":[42,144,231],"the":[43,67,85,93,96,99,106,114,145,152,159,172,179,192,196,202],"form":[44],"of":[45,69,95],"off-policy":[46,49,53,71,81,100,153,184,219],"traces.":[47],"Standard":[48],"supervise":[51],"against":[52],"data,":[54],"causing":[55],"instabilities":[56],"during":[57],"optimization.":[59],"We":[60,103],"introduce":[61],"PrefixRL,":[62],"condition":[65],"prefix":[68,101],"successful":[70],"run":[74],"to":[77,133,210],"complete":[78],"them,":[79],"side-stepping":[80],"instabilities.":[82],"PrefixRL":[83,107,170,214],"boosts":[84],"signal":[87],"problems":[90,131],"by":[91,155,205],"modulating":[92],"difficulty":[94],"problem":[97],"through":[98],"length.":[102],"prove":[104],"that":[105],"objective":[108,117],"is":[109,215],"not":[110],"only":[111,128],"consistent":[112],"with":[113,137,158],"standard":[115],"but":[118],"also":[119],"sample":[121],"efficient.":[122],"Empirically,":[123],"discover":[125],"back-generalization:":[126],"training":[127,174],"prefixed":[130],"generalizes":[132],"out-of-distribution":[134],"unprefixed":[135],"performance,":[136],"learned":[138],"strategies":[139],"often":[140],"differing":[141],"from":[142,223],"those":[143],"prefix.":[146],"In":[147],"our":[148],"experiments,":[149],"source":[151],"rejection":[156,198],"base":[160],"model,":[161],"creating":[162],"a":[163,224],"self-improvement":[164],"loop.":[165],"On":[166],"reaches":[171],"same":[173],"reward":[175,204],"2x":[176],"faster":[177],"than":[178],"strongest":[180],"baseline":[181],"(SFT":[182],"data":[185],"then":[186],"RL),":[187],"even":[188],"after":[189],"accounting":[190],"spent":[194],"initial":[197],"sampling,":[199],"increases":[201],"final":[203],"3x.":[206],"The":[207],"gains":[208],"transfer":[209],"held-out":[211],"benchmarks,":[212],"still":[216],"effective":[217],"when":[218],"derived":[222],"different":[225],"model":[226],"family,":[227],"validating":[228],"its":[229],"flexibility":[230],"practical":[232],"settings.":[233]},"counts_by_year":[],"updated_date":"2026-08-16T07:02:28.622633","created_date":"2026-01-28T00:00:00"}