{"id":"https://openalex.org/W7128714544","doi":"https://doi.org/10.48550/arxiv.2602.10238","title":"Learning to Evict from Key-Value Cache","display_name":"Learning to Evict from Key-Value Cache","publication_year":2026,"publication_date":"2026-02-10","ids":{"openalex":"https://openalex.org/W7128714544","doi":"https://doi.org/10.48550/arxiv.2602.10238"},"language":null,"primary_location":{"id":"pmh:doi:10.48550/arxiv.2602.10238","is_oa":true,"landing_page_url":null,"pdf_url":null,"source":{"id":"https://openalex.org/S4406922384","display_name":"Open MIND","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"publisher-specific-oa","license_id":"https://openalex.org/licenses/publisher-specific-oa","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Article"},"type":"article","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":null,"any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Moschella, Luca","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Moschella, Luca","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5125715142","display_name":"Laura Manduchi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Manduchi, Laura","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5026399398","display_name":"Ozan \u015eener","orcid":"https://orcid.org/0000-0002-1542-7547"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sener, Ozan","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":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.17002559,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"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":0.12489999830722809,"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":0.12489999830722809,"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":0.08860000222921371,"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/T13702","display_name":"Machine Learning in Healthcare","score":0.07699999958276749,"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/cache","display_name":"Cache","score":0.6685000061988831},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.6055999994277954},{"id":"https://openalex.org/keywords/benchmark","display_name":"Benchmark (surveying)","score":0.5465999841690063},{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.5404000282287598},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.5134999752044678},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.486299991607666},{"id":"https://openalex.org/keywords/scalability","display_name":"Scalability","score":0.4839000105857849},{"id":"https://openalex.org/keywords/deep-learning","display_name":"Deep learning","score":0.3871000111103058}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8079000115394592},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.6685000061988831},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.6055999994277954},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.5465999841690063},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.5404000282287598},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.5134999752044678},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.5055999755859375},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.486299991607666},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.4839000105857849},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4794999957084656},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.3871000111103058},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.36649999022483826},{"id":"https://openalex.org/C38556500","wikidata":"https://www.wikidata.org/wiki/Q13404475","display_name":"Cache algorithms","level":4,"score":0.32429999113082886},{"id":"https://openalex.org/C189783530","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"CPU cache","level":3,"score":0.3239000141620636},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.3140999972820282},{"id":"https://openalex.org/C187691185","wikidata":"https://www.wikidata.org/wiki/Q2020720","display_name":"Grid","level":2,"score":0.31029999256134033},{"id":"https://openalex.org/C774472","wikidata":"https://www.wikidata.org/wiki/Q6760393","display_name":"Margin (machine learning)","level":2,"score":0.2906000018119812},{"id":"https://openalex.org/C189430467","wikidata":"https://www.wikidata.org/wiki/Q7293293","display_name":"Ranking (information retrieval)","level":2,"score":0.2815000116825104},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.27320000529289246},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.2694000005722046},{"id":"https://openalex.org/C159877910","wikidata":"https://www.wikidata.org/wiki/Q2202883","display_name":"Autoregressive model","level":2,"score":0.26759999990463257},{"id":"https://openalex.org/C164226766","wikidata":"https://www.wikidata.org/wiki/Q7293202","display_name":"Rank (graph theory)","level":2,"score":0.2621000111103058},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.2565000057220459}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:doi:10.48550/arxiv.2602.10238","is_oa":true,"landing_page_url":null,"pdf_url":null,"source":{"id":"https://openalex.org/S4406922384","display_name":"Open MIND","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"publisher-specific-oa","license_id":"https://openalex.org/licenses/publisher-specific-oa","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Article"},{"id":"doi:10.48550/arxiv.2602.10238","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2602.10238","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:doi:10.48550/arxiv.2602.10238","is_oa":true,"landing_page_url":null,"pdf_url":null,"source":{"id":"https://openalex.org/S4406922384","display_name":"Open MIND","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"publisher-specific-oa","license_id":"https://openalex.org/licenses/publisher-specific-oa","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Article"},"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":{"The":[0],"growing":[1],"size":[2],"of":[3,18,89,125],"Large":[4],"Language":[5],"Models":[6],"(LLMs)":[7],"makes":[8],"efficient":[9],"inference":[10],"challenging,":[11],"primarily":[12],"due":[13],"to":[14,69,135,153,188,198],"the":[15,19,123,126,136,148,157],"memory":[16],"demands":[17],"autoregressive":[20],"Key-Value":[21],"(KV)":[22],"cache.":[23],"Existing":[24],"eviction":[25,61,110],"or":[26,38,139],"compression":[27],"methods":[28],"reduce":[29],"cost":[30],"but":[31],"rely":[32],"on":[33,95,147,169],"heuristics,":[34],"such":[35],"as":[36,45,62],"recency":[37],"past":[39],"attention":[40],"scores,":[41],"which":[42],"serve":[43],"only":[44,100],"indirect":[46],"proxies":[47],"for":[48,76,209],"a":[49,63,87,108,114,204],"token's":[50],"future":[51,77,119,200],"utility":[52,202],"and":[53,102,156,187,206],"introduce":[54,83],"computational":[55],"overhead.":[56],"We":[57],"reframe":[58],"KV":[59,84,211],"cache":[60,130,212],"reinforcement":[64],"learning":[65,68,197],"(RL)":[66],"problem:":[67],"rank":[70],"tokens":[71],"by":[72,113],"their":[73],"predicted":[74],"usefulness":[75],"decoding.":[78],"To":[79],"this":[80],"end,":[81],"we":[82],"Policy":[85],"(KVP),":[86],"framework":[88],"lightweight":[90],"per-head":[91],"RL":[92],"agents":[93],"trained":[94],"pre-computed":[96],"generation":[97],"traces":[98],"using":[99],"key":[101],"value":[103],"vectors.":[104],"Each":[105],"agent":[106],"learns":[107],"specialized":[109],"policy":[111],"guided":[112],"holistic":[115],"reward,":[116],"derived":[117],"from":[118],"utility,":[120],"that":[121,180,196],"evaluates":[122],"quality":[124],"ranking":[127],"across":[128,143],"all":[129],"budgets,":[131],"requiring":[132],"no":[133],"modifications":[134],"underlying":[137],"LLM":[138],"additional":[140],"inference.":[141],"Evaluated":[142],"two":[144],"model":[145],"families":[146],"long-context":[149],"benchmark":[150,160],"RULER":[151],"(up":[152],"128K":[154],"tokens)":[155],"multi-turn":[158],"dialogue":[159],"OASST2-4k,":[161],"KVP":[162,181],"significantly":[163],"outperforms":[164],"strong":[165],"baselines.":[166],"Zero-shot":[167],"tests":[168],"standard":[170],"downstream":[171],"tasks":[172],"(BoolQ,":[173],"LongBench":[174],"passage":[175],"retrieval,":[176],"GovReport)":[177],"further":[178],"show":[179],"generalizes":[182],"beyond":[183],"its":[184],"training":[185],"distribution":[186],"considerably":[189],"longer":[190],"sequence":[191],"lengths.":[192],"These":[193],"results":[194],"demonstrate":[195],"predict":[199],"token":[201],"is":[203],"powerful":[205],"scalable":[207],"paradigm":[208],"adaptive":[210],"management.":[213]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-02-13T00:00:00"}