{"id":"https://openalex.org/W4417051721","doi":"https://doi.org/10.1109/iccv51701.2025.02221","title":"AirCache: Activating Inter-Modal Relevancy KV Cache Compression for Efficient Large Vision-Language Model Inference","display_name":"AirCache: Activating Inter-Modal Relevancy KV Cache Compression for Efficient Large Vision-Language Model Inference","publication_year":2025,"publication_date":"2025-10-19","ids":{"openalex":"https://openalex.org/W4417051721","doi":"https://doi.org/10.1109/iccv51701.2025.02221"},"language":"en","primary_location":{"id":"doi:10.1109/iccv51701.2025.02221","is_oa":false,"landing_page_url":"https://doi.org/10.1109/iccv51701.2025.02221","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 IEEE/CVF International Conference on Computer Vision (ICCV)","raw_type":"proceedings-article"},"type":"article","indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2503.23956","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5100766400","display_name":"Kai Huang","orcid":"https://orcid.org/0009-0000-1735-1788"},"institutions":[{"id":"https://openalex.org/I4210095624","display_name":"Alibaba Group (United States)","ror":"https://ror.org/00rn0m335","country_code":"US","type":"company","lineage":["https://openalex.org/I4210095624","https://openalex.org/I45928872"]}],"countries":["US"],"is_corresponding":true,"raw_author_name":"Kai Huang","raw_affiliation_strings":["Alibaba Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Alibaba Group","institution_ids":["https://openalex.org/I4210095624"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5054034870","display_name":"Hao Zou","orcid":null},"institutions":[{"id":"https://openalex.org/I4210095624","display_name":"Alibaba Group (United States)","ror":"https://ror.org/00rn0m335","country_code":"US","type":"company","lineage":["https://openalex.org/I4210095624","https://openalex.org/I45928872"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Hao Zou","raw_affiliation_strings":["Alibaba Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Alibaba Group","institution_ids":["https://openalex.org/I4210095624"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5022002286","display_name":"Bochen Wang","orcid":null},"institutions":[{"id":"https://openalex.org/I4210095624","display_name":"Alibaba Group (United States)","ror":"https://ror.org/00rn0m335","country_code":"US","type":"company","lineage":["https://openalex.org/I4210095624","https://openalex.org/I45928872"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Bochen Wang","raw_affiliation_strings":["Alibaba Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Alibaba Group","institution_ids":["https://openalex.org/I4210095624"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100355602","display_name":"Xi Ye","orcid":"https://orcid.org/0000-0003-0331-7946"},"institutions":[{"id":"https://openalex.org/I4210095624","display_name":"Alibaba Group (United States)","ror":"https://ror.org/00rn0m335","country_code":"US","type":"company","lineage":["https://openalex.org/I4210095624","https://openalex.org/I45928872"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Ye Xi","raw_affiliation_strings":["Alibaba Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Alibaba Group","institution_ids":["https://openalex.org/I4210095624"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5037768390","display_name":"Zhen Xie","orcid":"https://orcid.org/0000-0003-3516-2192"},"institutions":[{"id":"https://openalex.org/I4210095624","display_name":"Alibaba Group (United States)","ror":"https://ror.org/00rn0m335","country_code":"US","type":"company","lineage":["https://openalex.org/I4210095624","https://openalex.org/I45928872"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zhen Xie","raw_affiliation_strings":["Alibaba Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Alibaba Group","institution_ids":["https://openalex.org/I4210095624"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5112080886","display_name":"Hao Wang","orcid":"https://orcid.org/0009-0003-4710-9249"},"institutions":[{"id":"https://openalex.org/I4210095624","display_name":"Alibaba Group (United States)","ror":"https://ror.org/00rn0m335","country_code":"US","type":"company","lineage":["https://openalex.org/I4210095624","https://openalex.org/I45928872"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Hao Wang","raw_affiliation_strings":["Alibaba Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Alibaba Group","institution_ids":["https://openalex.org/I4210095624"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":6,"corresponding_author_ids":["https://openalex.org/A5100766400"],"corresponding_institution_ids":["https://openalex.org/I4210095624"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.36630277,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"23958","last_page":"23967"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.9254999756813049,"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"}},"topics":[{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.9254999756813049,"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.014800000004470348,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.013299999758601189,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/cache","display_name":"Cache","score":0.661300003528595},{"id":"https://openalex.org/keywords/redundancy","display_name":"Redundancy (engineering)","score":0.5120000243186951},{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.4652999937534332},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.42980000376701355},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.42829999327659607},{"id":"https://openalex.org/keywords/decoding-methods","display_name":"Decoding methods","score":0.41359999775886536},{"id":"https://openalex.org/keywords/latency","display_name":"Latency (audio)","score":0.36809998750686646}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8335000276565552},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.661300003528595},{"id":"https://openalex.org/C152124472","wikidata":"https://www.wikidata.org/wiki/Q1204361","display_name":"Redundancy (engineering)","level":2,"score":0.5120000243186951},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.4652999937534332},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.42980000376701355},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.42829999327659607},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.41359999775886536},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.36809998750686646},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3668000102043152},{"id":"https://openalex.org/C189783530","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"CPU cache","level":3,"score":0.3479999899864197},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.3231000006198883},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.2978000044822693},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2955000102519989},{"id":"https://openalex.org/C38556500","wikidata":"https://www.wikidata.org/wiki/Q13404475","display_name":"Cache algorithms","level":4,"score":0.29330000281333923},{"id":"https://openalex.org/C2778751112","wikidata":"https://www.wikidata.org/wiki/Q835016","display_name":"Window (computing)","level":2,"score":0.2718999981880188},{"id":"https://openalex.org/C179518139","wikidata":"https://www.wikidata.org/wiki/Q5140297","display_name":"Coding (social sciences)","level":2,"score":0.2637999951839447},{"id":"https://openalex.org/C183322885","wikidata":"https://www.wikidata.org/wiki/Q17007702","display_name":"Context model","level":3,"score":0.26339998841285706},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.258899986743927},{"id":"https://openalex.org/C36464697","wikidata":"https://www.wikidata.org/wiki/Q451553","display_name":"Visualization","level":2,"score":0.25699999928474426},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.2508000135421753}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.1109/iccv51701.2025.02221","is_oa":false,"landing_page_url":"https://doi.org/10.1109/iccv51701.2025.02221","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 IEEE/CVF International Conference on Computer Vision (ICCV)","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2503.23956","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2503.23956","pdf_url":"https://arxiv.org/pdf/2503.23956","source":{"id":"https://openalex.org/S4393918464","display_name":"ArXiv.org","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2503.23956","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2503.23956","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":"pmh:oai:arXiv.org:2503.23956","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2503.23956","pdf_url":"https://arxiv.org/pdf/2503.23956","source":{"id":"https://openalex.org/S4393918464","display_name":"ArXiv.org","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"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":{"Recent":[0],"advancements":[1],"in":[2,20,88,122],"Large":[3],"Visual":[4],"Language":[5],"Models":[6],"(LVLMs)":[7],"have":[8],"gained":[9],"significant":[10],"attention":[11,78],"due":[12],"to":[13,39,160,177,195],"their":[14],"remarkable":[15],"reasoning":[16],"capabilities":[17],"and":[18,30,73,150,168,201],"proficiency":[19],"generalization.":[21],"However,":[22],"processing":[23],"a":[24,54],"large":[25],"number":[26],"of":[27,80,119,152,185,204],"visual":[28,72,90,120,186],"tokens":[29,75,96],"generating":[31],"long-context":[32],"outputs":[33],"impose":[34],"substantial":[35],"computational":[36],"overhead,":[37],"leading":[38],"excessive":[40],"demands":[41],"for":[42,115],"key-value":[43],"(KV)":[44],"cache.":[45],"To":[46],"address":[47],"this":[48],"critical":[49],"bottleneck,":[50],"we":[51,109,137],"propose":[52],"AirCache,":[53],"novel":[55],"KV":[56,124,187],"cache":[57,180,208],"compression":[58],"method":[59,173,213],"aimed":[60],"at":[61],"accelerating":[62,102],"LVLMs":[63,167],"inference.":[64],"This":[65],"work":[66],"systematically":[67],"investigates":[68],"the":[69,77,117,123,148,178],"correlations":[70],"between":[71],"textual":[74],"within":[76],"mechanisms":[79],"LVLMs.":[81],"Our":[82],"empirical":[83],"analysis":[84],"reveals":[85],"considerable":[86],"redundancy":[87],"cached":[89],"tokens,":[91],"wherein":[92],"strategically":[93],"eliminating":[94],"these":[95,107],"preserves":[97],"model":[98],"performance":[99,176,216],"while":[100,181],"significantly":[101],"context":[103],"generation.":[104],"Inspired":[105],"by":[106,193],"findings,":[108],"introduce":[110],"an":[111,139],"elite":[112],"observation":[113],"window":[114],"assessing":[116],"importance":[118,154],"components":[121],"cache,":[125,188],"focusing":[126],"on":[127,147],"stable":[128],"inter-modal":[129],"relevancy":[130],"modeling":[131],"with":[132],"enhanced":[133],"multi-perspective":[134],"consistency.":[135],"Additionally,":[136],"develop":[138],"adaptive":[140],"layer-wise":[141],"budget":[142],"allocation":[143],"strategy":[144],"that":[145,171],"capitalizes":[146],"strength":[149],"skewness":[151],"token":[153],"distribution,":[155],"showcasing":[156],"superior":[157],"efficiency":[158],"compared":[159],"uniform":[161],"allocation.":[162],"Comprehensive":[163],"evaluations":[164],"across":[165,197],"multiple":[166],"benchmarks":[169],"demonstrate":[170],"our":[172,212],"achieves":[174],"comparable":[175],"full":[179],"retaining":[182],"only":[183],"10%":[184],"thereby":[189],"reducing":[190],"decoding":[191],"latency":[192],"29%":[194],"66%":[196],"various":[198],"batch":[199],"size":[200],"prompt":[202],"length":[203],"inputs.":[205],"Notably,":[206],"as":[207],"retention":[209],"rates":[210],"decrease,":[211],"exhibits":[214],"increasing":[215],"advantages":[217],"over":[218],"existing":[219],"approaches.":[220]},"counts_by_year":[],"updated_date":"2026-05-06T06:03:25.996018","created_date":"2025-10-10T00:00:00"}