{"id":"https://openalex.org/W7133558784","doi":"https://doi.org/10.1109/tcasai.2026.3670075","title":"Asymmetric KV Cache Compression Using State-Aware Sparsity and Quantization","display_name":"Asymmetric KV Cache Compression Using State-Aware Sparsity and Quantization","publication_year":2026,"publication_date":"2026-03-04","ids":{"openalex":"https://openalex.org/W7133558784","doi":"https://doi.org/10.1109/tcasai.2026.3670075"},"language":null,"primary_location":{"id":"doi:10.1109/tcasai.2026.3670075","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tcasai.2026.3670075","pdf_url":null,"source":{"id":"https://openalex.org/S4404675360","display_name":"IEEE transactions on circuits and systems for artificial intelligence.","issn_l":"2996-6647","issn":["2996-6647"],"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":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"IEEE Transactions on Circuits and Systems for Artificial Intelligence","raw_type":"journal-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5028977510","display_name":"Myeongji Yun","orcid":"https://orcid.org/0009-0002-0538-6427"},"institutions":[{"id":"https://openalex.org/I27837315","display_name":"University of Michigan","ror":"https://ror.org/00jmfr291","country_code":"US","type":"education","lineage":["https://openalex.org/I27837315"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Myeongji Yun","raw_affiliation_strings":["University of Michigan"],"raw_orcid":"https://orcid.org/0009-0008-5340-9420","affiliations":[{"raw_affiliation_string":"University of Michigan","institution_ids":["https://openalex.org/I27837315"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5064486180","display_name":"Dongyun Kam","orcid":"https://orcid.org/0000-0002-8542-1845"},"institutions":[{"id":"https://openalex.org/I48566637","display_name":"Ulsan National Institute of Science and Technology","ror":"https://ror.org/017cjz748","country_code":"KR","type":"education","lineage":["https://openalex.org/I48566637"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Dongyun Kam","raw_affiliation_strings":["Ulsan National Institute of Science and Technology (UNIST)"],"raw_orcid":"https://orcid.org/0000-0002-8542-1845","affiliations":[{"raw_affiliation_string":"Ulsan National Institute of Science and Technology (UNIST)","institution_ids":["https://openalex.org/I48566637"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5050310538","display_name":"Priyadarshini Panda","orcid":"https://orcid.org/0000-0002-4167-6782"},"institutions":[{"id":"https://openalex.org/I1174212","display_name":"University of Southern California","ror":"https://ror.org/03taz7m60","country_code":"US","type":"education","lineage":["https://openalex.org/I1174212"]},{"id":"https://openalex.org/I2800817003","display_name":"California Southern University","ror":"https://ror.org/058zz0t50","country_code":"US","type":"education","lineage":["https://openalex.org/I2800817003"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Priyadarshini Panda","raw_affiliation_strings":["University of Southern California"],"raw_orcid":"https://orcid.org/0000-0002-4167-6782","affiliations":[{"raw_affiliation_string":"University of Southern California","institution_ids":["https://openalex.org/I1174212","https://openalex.org/I2800817003"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5126436384","display_name":"Youngjoo Lee","orcid":null},"institutions":[{"id":"https://openalex.org/I157485424","display_name":"Korea Advanced Institute of Science and Technology","ror":"https://ror.org/05apxxy63","country_code":"KR","type":"education","lineage":["https://openalex.org/I157485424"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Youngjoo Lee","raw_affiliation_strings":["Korea Advanced Institute of Science and Technology (KAIST)"],"raw_orcid":"https://orcid.org/0000-0002-2467-8276","affiliations":[{"raw_affiliation_string":"Korea Advanced Institute of Science and Technology (KAIST)","institution_ids":["https://openalex.org/I157485424"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":5,"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":0.18019117,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":"3","issue":"4","first_page":"246","last_page":"259"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.3637999892234802,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.3637999892234802,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"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/T11478","display_name":"Caching and Content Delivery","score":0.17960000038146973,"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/T11269","display_name":"Algorithms and Data Compression","score":0.05510000139474869,"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/quantization","display_name":"Quantization (signal processing)","score":0.5006999969482422},{"id":"https://openalex.org/keywords/compression","display_name":"Compression (physics)","score":0.44200000166893005},{"id":"https://openalex.org/keywords/data-compression","display_name":"Data compression","score":0.4246000051498413},{"id":"https://openalex.org/keywords/vector-quantization","display_name":"Vector quantization","score":0.3416999876499176},{"id":"https://openalex.org/keywords/cache","display_name":"Cache","score":0.3025999963283539}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5867000222206116},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.5465999841690063},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.5006999969482422},{"id":"https://openalex.org/C180016635","wikidata":"https://www.wikidata.org/wiki/Q2712821","display_name":"Compression (physics)","level":2,"score":0.44200000166893005},{"id":"https://openalex.org/C78548338","wikidata":"https://www.wikidata.org/wiki/Q2493","display_name":"Data compression","level":2,"score":0.4246000051498413},{"id":"https://openalex.org/C199833920","wikidata":"https://www.wikidata.org/wiki/Q612536","display_name":"Vector quantization","level":2,"score":0.3416999876499176},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.3025999963283539},{"id":"https://openalex.org/C81081738","wikidata":"https://www.wikidata.org/wiki/Q55542","display_name":"Lossless compression","level":3,"score":0.28600001335144043},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.2703000009059906},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.26330000162124634}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/tcasai.2026.3670075","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tcasai.2026.3670075","pdf_url":null,"source":{"id":"https://openalex.org/S4404675360","display_name":"IEEE transactions on circuits and systems for artificial intelligence.","issn_l":"2996-6647","issn":["2996-6647"],"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":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"IEEE Transactions on Circuits and Systems for Artificial Intelligence","raw_type":"journal-article"}],"best_oa_location":null,"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":{"To":[0],"enable":[1],"the":[2,42,51,63,118,130,136,151],"efficient":[3],"deployment":[4],"of":[5,65,107,132,153],"Large":[6],"Language":[7],"Models":[8],"(LLMs)":[9],"on":[10,143],"resource-constrained":[11],"devices,":[12],"recent":[13],"studies":[14],"have":[15],"explored":[16],"Key-Value":[17],"(KV)":[18],"Cache":[19,85,123],"compression,":[20],"such":[21],"as":[22,39],"quantization":[23,92],"and":[24,36,47,50,56,97,110,121,146,162,183,188,201,211],"sparsity-based":[25,54,94],"methods.":[26],"However,":[27],"prior":[28],"works":[29,60],"suffer":[30],"from":[31],"limitations":[32],"in":[33,208],"both":[34,209],"accuracy":[35,119,176],"memory":[37,189,203],"efficiency,":[38],"they":[40],"overlook":[41],"structural":[43,112],"differences":[44],"between":[45,53],"Key":[46],"Value":[48],"states":[49],"interaction":[52],"compression":[55,86,95,126],"quantization.":[57],"Moreover,":[58],"previous":[59],"often":[61],"underexplored":[62],"variability":[64],"generated":[66],"token":[67],"lengths":[68,207],"across":[69,205],"requests,":[70],"leading":[71],"to":[72,100,129,174],"inconsistent":[73],"performance.To":[74],"address":[75],"these":[76],"limitations,":[77],"we":[78,115,149],"present":[79],"a":[80,164],"new":[81,165],"asymmetric,":[82],"state-aware":[83],"KV":[84,122],"pipeline.":[87],"The":[88],"proposed":[89],"pipeline":[90],"tailors":[91],"granularity,":[93],"type,":[96],"sparsity":[98],"level":[99],"each":[101],"state":[102],"guided":[103],"by":[104],"our":[105,144,195],"analysis":[106],"methodological":[108],"characteristics":[109],"state-specific":[111],"properties.":[113],"Furthermore,":[114],"demonstrate":[116,193],"that":[117,194],"degradation":[120],"footprint":[124],"under":[125],"are":[127],"sensitive":[128],"ordering":[131],"different":[133],"operations,":[134],"highlighting":[135],"need":[137],"for":[138,167],"an":[139],"order-aware":[140],"design.":[141],"Building":[142],"state-":[145],"order-specific":[147],"insights,":[148],"assess":[150],"influence":[152],"individual":[154],"channels":[155],"before":[156],"applying":[157],"Rotary":[158],"Positional":[159],"Embedding":[160],"(RoPE),":[161],"define":[163],"criterion":[166],"structured":[168],"channel":[169],"pruning,":[170],"enabling":[171],"pre-RoPE":[172],"pruning":[173],"mitigate":[175],"loss.":[177],"We":[178],"also":[179],"incorporate":[180],"parameter-reusing":[181],"techniques":[182],"avoid":[184],"extra":[185],"decoding-time":[186],"computation":[187],"overhead.":[190],"Experimental":[191],"results":[192],"approach":[196],"achieves":[197],"superior":[198],"accuracy-memory":[199],"trade-offs":[200],"consistent":[202],"savings":[204],"sequence":[206],"quantized":[210],"non-quantized":[212],"settings.":[213]},"counts_by_year":[],"updated_date":"2026-08-01T09:00:35.917206","created_date":"2026-03-05T00:00:00"}