{"id":"https://openalex.org/W7164904434","doi":"https://doi.org/10.48550/arxiv.2606.17024","title":"ExpRL: Exploratory RL for LLM Mid-Training","display_name":"ExpRL: Exploratory RL for LLM Mid-Training","publication_year":2026,"publication_date":"2026-06-15","ids":{"openalex":"https://openalex.org/W7164904434","doi":"https://doi.org/10.48550/arxiv.2606.17024"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.17024","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.17024","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.2606.17024","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5023185054","display_name":"Violet Xiang","orcid":"https://orcid.org/0000-0002-1584-6231"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiang, Violet","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138738769","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/A5117274429","display_name":"Chase Blagden","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Blagden, Chase","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138752345","display_name":"Nick Haber","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Haber, Nick","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5138723657","display_name":"Aviral Kumar","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kumar, Aviral","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":0.5152999758720398,"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":0.5152999758720398,"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":0.1534000039100647,"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/T10462","display_name":"Reinforcement Learning in Robotics","score":0.15230000019073486,"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/reinforcement-learning","display_name":"Reinforcement learning","score":0.5971999764442444},{"id":"https://openalex.org/keywords/rubric","display_name":"Rubric","score":0.5773000121116638},{"id":"https://openalex.org/keywords/heuristics","display_name":"Heuristics","score":0.47110000252723694},{"id":"https://openalex.org/keywords/construct","display_name":"Construct (python library)","score":0.47099998593330383},{"id":"https://openalex.org/keywords/initialization","display_name":"Initialization","score":0.46560001373291016},{"id":"https://openalex.org/keywords/knowledge-base","display_name":"Knowledge base","score":0.35760000348091125},{"id":"https://openalex.org/keywords/trace","display_name":"TRACE (psycholinguistics)","score":0.35519999265670776}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7243000268936157},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.5971999764442444},{"id":"https://openalex.org/C111640148","wikidata":"https://www.wikidata.org/wiki/Q847349","display_name":"Rubric","level":2,"score":0.5773000121116638},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5454999804496765},{"id":"https://openalex.org/C127705205","wikidata":"https://www.wikidata.org/wiki/Q5748245","display_name":"Heuristics","level":2,"score":0.47110000252723694},{"id":"https://openalex.org/C2780801425","wikidata":"https://www.wikidata.org/wiki/Q5164392","display_name":"Construct (python library)","level":2,"score":0.47099998593330383},{"id":"https://openalex.org/C114466953","wikidata":"https://www.wikidata.org/wiki/Q6034165","display_name":"Initialization","level":2,"score":0.46560001373291016},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3837999999523163},{"id":"https://openalex.org/C4554734","wikidata":"https://www.wikidata.org/wiki/Q593744","display_name":"Knowledge base","level":2,"score":0.35760000348091125},{"id":"https://openalex.org/C75291252","wikidata":"https://www.wikidata.org/wiki/Q1315756","display_name":"TRACE (psycholinguistics)","level":2,"score":0.35519999265670776},{"id":"https://openalex.org/C75553542","wikidata":"https://www.wikidata.org/wiki/Q178161","display_name":"A priori and a posteriori","level":2,"score":0.3278999924659729},{"id":"https://openalex.org/C195344581","wikidata":"https://www.wikidata.org/wiki/Q2555318","display_name":"Automated reasoning","level":2,"score":0.3068999946117401},{"id":"https://openalex.org/C81444415","wikidata":"https://www.wikidata.org/wiki/Q7243535","display_name":"Priming (agriculture)","level":3,"score":0.2865999937057495},{"id":"https://openalex.org/C100660578","wikidata":"https://www.wikidata.org/wiki/Q18733","display_name":"Recall","level":2,"score":0.2847000062465668},{"id":"https://openalex.org/C159032336","wikidata":"https://www.wikidata.org/wiki/Q2488768","display_name":"Non-monotonic logic","level":2,"score":0.27219998836517334},{"id":"https://openalex.org/C42058472","wikidata":"https://www.wikidata.org/wiki/Q810214","display_name":"Base (topology)","level":2,"score":0.2623000144958496}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.17024","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.17024","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.2606.17024","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.17024","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":[{"id":"https://metadata.un.org/sdg/4","score":0.567343533039093,"display_name":"Quality Education"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Sparse":[0],"reward":[1],"reinforcement":[2],"learning":[3],"(RL)":[4],"has":[5],"become":[6],"a":[7,89,205],"standard":[8],"tool":[9],"for":[10,33,74,133,208],"improving":[11],"LLM":[12,148],"reasoning,":[13],"but":[14],"its":[15],"success":[16],"depends":[17],"critically":[18],"on":[19,37],"the":[20,24,60,123,142,151,156,222],"coverage":[21,71],"present":[22],"in":[23],"base":[25],"model.":[26],"In":[27],"practice,":[28],"models":[29],"are":[30,120],"often":[31,183],"primed":[32],"RL":[34,195],"through":[35],"\\emph{mid-training}":[36],"curated":[38],"reasoning":[39,136,153,177,190],"traces":[40],"that":[41,179,217],"teach":[42],"useful":[43,172],"primitive":[44,70],"skills":[45,82],"such":[46,69],"as":[47,107,116],"decomposition,":[48],"verification,":[49],"or":[50,162],"self-correction.":[51],"Although":[52],"effective,":[53],"this":[54],"strategy":[55],"requires":[56],"manually":[57],"specifying":[58],"what":[59],"model":[61],"should":[62],"learn,":[63],"and":[64,125,159,175,201,203],"it":[65],"remains":[66],"unclear":[67],"whether":[68],"is":[72],"enough":[73],"much":[75],"harder":[76],"problems,":[77],"which":[78],"require":[79],"combining":[80],"these":[81],"into":[83],"broader":[84],"solution":[85,158],"strategies.":[86],"We":[87],"study":[88],"more":[90],"automated":[91],"approach:":[92],"\\emph{RL-based":[93],"mid-training}":[94],"using":[95],"large":[96],"corpora":[97],"of":[98],"human-written":[99],"question-answer":[100],"data.":[101],"Rather":[102],"than":[103,197],"treating":[104],"reference":[105,157],"solutions":[106],"targets":[108],"to":[109,128,185],"imitate,":[110],"our":[111],"method,":[112],"ExpRL,":[113],"uses":[114],"them":[115],"\\emph{reward":[117],"scaffolds}:":[118],"references":[119],"hidden":[121],"from":[122,141],"policy":[124,139],"used":[126],"only":[127],"construct":[129],"problem-specific":[130],"grading":[131],"rubrics":[132],"judging":[134],"on-policy":[135],"traces.":[137],"The":[138],"samples":[140],"original":[143,223],"problem":[144],"prompt,":[145],"while":[146],"an":[147],"judge":[149],"compares":[150],"sampled":[152],"trace":[154],"against":[155],"assigns":[160],"outcome-level":[161],"process-level":[163],"dense":[164],"rewards.":[165],"This":[166],"lets":[167],"ExpRL":[168,192,218],"reinforce":[169],"partial":[170],"progress,":[171],"intermediate":[173],"reductions,":[174],"productive":[176],"behaviors":[178],"sparse":[180],"final-answer":[181],"rewards":[182],"fail":[184],"upweight.":[186],"On":[187],"challenging":[188],"math":[189],"tasks,":[191],"yields":[193],"stronger":[194],"priming":[196],"SFT,":[198],"sparse-reward":[199,210],"GRPO,":[200],"self-distillation,":[202],"provides":[204],"better":[206],"initialization":[207],"subsequent":[209],"RL.":[211],"Additional":[212],"mixed-domain":[213],"experiments":[214],"further":[215],"suggest":[216],"can":[219],"extend":[220],"beyond":[221],"math-only":[224],"setting.":[225]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-17T00:00:00"}