{"id":"https://openalex.org/W7154120085","doi":"https://doi.org/10.48550/arxiv.2604.08706","title":"Efficient RL Training for LLMs with Experience Replay","display_name":"Efficient RL Training for LLMs with Experience Replay","publication_year":2026,"publication_date":"2026-04-09","ids":{"openalex":"https://openalex.org/W7154120085","doi":"https://doi.org/10.48550/arxiv.2604.08706"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.08706","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.08706","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2604.08706","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5070265689","display_name":"Charles Arnal","orcid":"https://orcid.org/0000-0002-3306-0574"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Arnal, Charles","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5059273291","display_name":"Vivien Cabannes","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cabannes, Vivien","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133515955","display_name":"Taco Cohen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cohen, Taco","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5051933322","display_name":"Julia Kempe","orcid":"https://orcid.org/0009-0006-1984-0060"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kempe, Julia","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5061574062","display_name":"R\u00e9mi Munos","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Munos, Remi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":5,"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/T12127","display_name":"Software System Performance and Reliability","score":0.07660000026226044,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/T12127","display_name":"Software System Performance and Reliability","score":0.07660000026226044,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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":0.07440000027418137,"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/T12761","display_name":"Data Stream Mining Techniques","score":0.06509999930858612,"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/training","display_name":"Training (meteorology)","score":0.6291999816894531},{"id":"https://openalex.org/keywords/sample","display_name":"Sample (material)","score":0.579200029373169},{"id":"https://openalex.org/keywords/reuse","display_name":"Reuse","score":0.5435000061988831},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.504800021648407},{"id":"https://openalex.org/keywords/sampling","display_name":"Sampling (signal processing)","score":0.4293999969959259},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.3610000014305115},{"id":"https://openalex.org/keywords/diversity","display_name":"Diversity (politics)","score":0.3547999858856201}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7251999974250793},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.6291999816894531},{"id":"https://openalex.org/C198531522","wikidata":"https://www.wikidata.org/wiki/Q485146","display_name":"Sample (material)","level":2,"score":0.579200029373169},{"id":"https://openalex.org/C206588197","wikidata":"https://www.wikidata.org/wiki/Q846574","display_name":"Reuse","level":2,"score":0.5435000061988831},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.504800021648407},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.4293999969959259},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.3610000014305115},{"id":"https://openalex.org/C2781316041","wikidata":"https://www.wikidata.org/wiki/Q1230584","display_name":"Diversity (politics)","level":2,"score":0.3547999858856201},{"id":"https://openalex.org/C136197465","wikidata":"https://www.wikidata.org/wiki/Q1729295","display_name":"Variety (cybernetics)","level":2,"score":0.35249999165534973},{"id":"https://openalex.org/C112930515","wikidata":"https://www.wikidata.org/wiki/Q4389547","display_name":"Risk analysis (engineering)","level":1,"score":0.3492000102996826},{"id":"https://openalex.org/C195094911","wikidata":"https://www.wikidata.org/wiki/Q14167904","display_name":"Process management","level":1,"score":0.3375999927520752},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3174000084400177},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3001999855041504},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.2992999851703644},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.2973000109195709},{"id":"https://openalex.org/C184356942","wikidata":"https://www.wikidata.org/wiki/Q830382","display_name":"Best practice","level":2,"score":0.2802000045776367},{"id":"https://openalex.org/C49774154","wikidata":"https://www.wikidata.org/wiki/Q131765","display_name":"Multimedia","level":1,"score":0.2587999999523163},{"id":"https://openalex.org/C2522767166","wikidata":"https://www.wikidata.org/wiki/Q2374463","display_name":"Data science","level":1,"score":0.25609999895095825},{"id":"https://openalex.org/C206345919","wikidata":"https://www.wikidata.org/wiki/Q20380951","display_name":"Resource (disambiguation)","level":2,"score":0.25589999556541443},{"id":"https://openalex.org/C3019252630","wikidata":"https://www.wikidata.org/wiki/Q6549547","display_name":"Limited resources","level":2,"score":0.2531999945640564}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.08706","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.08706","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"article"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2604.08706","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.08706","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"article"},"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":{"While":[0],"Experience":[1],"Replay":[2],"-":[3,16,109,116],"the":[4,33,64,76],"practice":[5],"of":[6,57,80],"storing":[7],"rollouts":[8],"and":[9,75,110],"reusing":[10],"them":[11],"multiple":[12],"times":[13],"during":[14],"training":[15],"is":[17,40,88,92],"a":[18,54,68,98],"foundational":[19],"technique":[20],"in":[21,28,111],"general":[22],"RL,":[23],"it":[24],"remains":[25],"largely":[26],"unexplored":[27],"LLM":[29,61],"post-training":[30],"due":[31],"to":[32],"prevailing":[34],"belief":[35],"that":[36,84,97],"fresh,":[37],"on-policy":[38,86],"data":[39],"essential":[41],"for":[42,60],"high":[43,77],"performance.":[44],"In":[45],"this":[46,50],"work,":[47],"we":[48,95],"challenge":[49],"assumption.":[51],"We":[52,82],"present":[53],"systematic":[55],"study":[56],"replay":[58,100],"buffers":[59],"post-training,":[62],"formalizing":[63],"optimal":[65],"design":[66],"as":[67],"trade-off":[69],"between":[70],"staleness-induced":[71],"variance,":[72],"sample":[73],"diversity":[74],"computational":[78],"cost":[79],"generation.":[81],"show":[83,96],"strict":[85],"sampling":[87],"suboptimal":[89],"when":[90],"generation":[91],"expensive.":[93],"Empirically,":[94],"well-designed":[99],"buffer":[101],"can":[102],"drastically":[103],"reduce":[104],"inference":[105],"compute":[106],"without":[107],"degrading":[108],"some":[112],"cases":[113],"even":[114],"improving":[115],"final":[117],"model":[118],"performance,":[119],"while":[120],"preserving":[121],"policy":[122],"entropy.":[123]},"counts_by_year":[],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2026-04-14T00:00:00"}