{"id":"https://openalex.org/W7134946070","doi":"https://doi.org/10.48550/arxiv.2603.09229","title":"Flash-KMeans: Fast and Memory-Efficient Exact K-Means","display_name":"Flash-KMeans: Fast and Memory-Efficient Exact K-Means","publication_year":2026,"publication_date":"2026-03-10","ids":{"openalex":"https://openalex.org/W7134946070","doi":"https://doi.org/10.48550/arxiv.2603.09229"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2603.09229","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.09229","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.2603.09229","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5128757491","display_name":"Shuo Yang","orcid":null},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Yang, Shuo","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5041595650","display_name":"Haocheng Xi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xi, Haocheng","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128791297","display_name":"Yilong Zhao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao, Yilong","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128709453","display_name":"Muyang Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Muyang","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128767762","display_name":"Xiaoze Fan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Fan, Xiaoze","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128768763","display_name":"Jintao Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Jintao","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128742758","display_name":"Han Cai","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cai, Han","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128705133","display_name":"Yujun Lin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lin, Yujun","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128755282","display_name":"Xiuyu Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Xiuyu","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5047285420","display_name":"Kurt Keutzer","orcid":"https://orcid.org/0000-0003-3868-8501"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Keutzer, Kurt","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128794305","display_name":"Song Han","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Han, Song","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100953941","display_name":"Chenfeng Xu","orcid":"https://orcid.org/0009-0001-0357-6203"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xu, Chenfeng","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5128795547","display_name":"Ion Stoica","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Stoica, Ion","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":13,"corresponding_author_ids":["https://openalex.org/A5128757491"],"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/T12292","display_name":"Graph Theory and Algorithms","score":0.46140000224113464,"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/T12292","display_name":"Graph Theory and Algorithms","score":0.46140000224113464,"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/T10101","display_name":"Cloud Computing and Resource Management","score":0.2045000046491623,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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/T10317","display_name":"Advanced Database Systems and Queries","score":0.030300000682473183,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/bottleneck","display_name":"Bottleneck","score":0.7775999903678894},{"id":"https://openalex.org/keywords/speedup","display_name":"Speedup","score":0.5730000138282776},{"id":"https://openalex.org/keywords/embedding","display_name":"Embedding","score":0.5353999733924866},{"id":"https://openalex.org/keywords/preprocessor","display_name":"Preprocessor","score":0.5187000036239624},{"id":"https://openalex.org/keywords/bridge","display_name":"Bridge (graph theory)","score":0.4814000129699707},{"id":"https://openalex.org/keywords/memory-bandwidth","display_name":"Memory bandwidth","score":0.4657999873161316},{"id":"https://openalex.org/keywords/implementation","display_name":"Implementation","score":0.4645000100135803},{"id":"https://openalex.org/keywords/computation","display_name":"Computation","score":0.4535999894142151},{"id":"https://openalex.org/keywords/matrix-multiplication","display_name":"Matrix multiplication","score":0.39100000262260437}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8309999704360962},{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.7775999903678894},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.5730000138282776},{"id":"https://openalex.org/C41608201","wikidata":"https://www.wikidata.org/wiki/Q980509","display_name":"Embedding","level":2,"score":0.5353999733924866},{"id":"https://openalex.org/C34736171","wikidata":"https://www.wikidata.org/wiki/Q918333","display_name":"Preprocessor","level":2,"score":0.5187000036239624},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.5112000107765198},{"id":"https://openalex.org/C100776233","wikidata":"https://www.wikidata.org/wiki/Q2532492","display_name":"Bridge (graph theory)","level":2,"score":0.4814000129699707},{"id":"https://openalex.org/C188045654","wikidata":"https://www.wikidata.org/wiki/Q17148339","display_name":"Memory bandwidth","level":2,"score":0.4657999873161316},{"id":"https://openalex.org/C26713055","wikidata":"https://www.wikidata.org/wiki/Q245962","display_name":"Implementation","level":2,"score":0.4645000100135803},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.45730000734329224},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.4535999894142151},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.4020000100135803},{"id":"https://openalex.org/C17349429","wikidata":"https://www.wikidata.org/wiki/Q1049914","display_name":"Matrix multiplication","level":3,"score":0.39100000262260437},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.38260000944137573},{"id":"https://openalex.org/C169590947","wikidata":"https://www.wikidata.org/wiki/Q47506","display_name":"Compiler","level":2,"score":0.38119998574256897},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.3517000079154968},{"id":"https://openalex.org/C78766204","wikidata":"https://www.wikidata.org/wiki/Q555032","display_name":"Multi-core processor","level":2,"score":0.34950000047683716},{"id":"https://openalex.org/C207467116","wikidata":"https://www.wikidata.org/wiki/Q4385666","display_name":"Inverse","level":2,"score":0.34389999508857727},{"id":"https://openalex.org/C2776257435","wikidata":"https://www.wikidata.org/wiki/Q1576430","display_name":"Bandwidth (computing)","level":2,"score":0.3409000039100647},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.33959999680519104},{"id":"https://openalex.org/C84211073","wikidata":"https://www.wikidata.org/wiki/Q117879","display_name":"Floating point","level":2,"score":0.3084999918937683},{"id":"https://openalex.org/C2781357197","wikidata":"https://www.wikidata.org/wiki/Q5757597","display_name":"High memory","level":2,"score":0.30720001459121704},{"id":"https://openalex.org/C28719098","wikidata":"https://www.wikidata.org/wiki/Q44946","display_name":"Point (geometry)","level":2,"score":0.304500013589859},{"id":"https://openalex.org/C106487976","wikidata":"https://www.wikidata.org/wiki/Q685816","display_name":"Matrix (chemical analysis)","level":2,"score":0.3012000024318695},{"id":"https://openalex.org/C162319229","wikidata":"https://www.wikidata.org/wiki/Q175263","display_name":"Data structure","level":2,"score":0.2685000002384186},{"id":"https://openalex.org/C2777212361","wikidata":"https://www.wikidata.org/wiki/Q5127848","display_name":"Class (philosophy)","level":2,"score":0.26759999990463257},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.26489999890327454},{"id":"https://openalex.org/C136886441","wikidata":"https://www.wikidata.org/wiki/Q926129","display_name":"Normalization (sociology)","level":2,"score":0.2644999921321869},{"id":"https://openalex.org/C74912251","wikidata":"https://www.wikidata.org/wiki/Q6815727","display_name":"Memory footprint","level":2,"score":0.25589999556541443}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2603.09229","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.09229","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.2603.09229","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.09229","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":[{"id":"https://metadata.un.org/sdg/9","display_name":"Industry, innovation and infrastructure","score":0.5964937806129456}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"$k$-means":[0,46,59,131],"has":[1],"historically":[2],"been":[3],"positioned":[4],"primarily":[5],"as":[6,21,47],"an":[7,48,127,150,165],"offline":[8],"processing":[9],"primitive,":[10],"typically":[11],"used":[12],"for":[13,133],"dataset":[14],"organization":[15],"or":[16],"embedding":[17],"preprocessing":[18],"rather":[19,67],"than":[20,68],"a":[22,78],"first-class":[23],"component":[24],"in":[25,95],"online":[26,49,151],"systems.":[27],"In":[28],"this":[29,33,121],"work,":[30],"we":[31,124,179],"revisit":[32],"classical":[34],"algorithm":[35],"under":[36],"the":[37,73,84,89,101],"lens":[38],"of":[39,58,88],"modern":[40,134],"AI":[41],"system":[42,65],"design":[43],"and":[44,129,186,218,222],"enable":[45],"primitive.":[50],"We":[51],"point":[52],"out":[53],"that":[54,201],"existing":[55],"GPU":[56,135],"implementations":[57],"remain":[60],"fundamentally":[61],"bottlenecked":[62],"by":[63,108,114,220],"low-level":[64],"constraints":[66],"theoretical":[69],"algorithmic":[70],"complexity.":[71],"Specifically,":[72],"assignment":[74],"stage":[75,104],"suffers":[76],"from":[77],"severe":[79],"IO":[80],"bottleneck":[81],"due":[82],"to":[83,153,168,190,205],"massive":[85],"explicit":[86],"materialization":[87],"$N":[90],"\\times":[91],"K$":[92],"distance":[93,147],"matrix":[94],"High":[96],"Bandwidth":[97],"Memory":[98],"(HBM).":[99],"Simultaneously,":[100],"centroid":[102],"update":[103],"is":[105,228],"heavily":[106],"penalized":[107],"hardware-level":[109],"atomic":[110,171],"write":[111],"contention":[112],"caused":[113],"irregular,":[115],"scatter-style":[116],"token":[117],"aggregations.":[118],"To":[119],"bridge":[120],"performance":[122],"gap,":[123],"propose":[125],"flash-kmeans,":[126],"IO-aware":[128],"contention-free":[130],"implementation":[132],"workloads.":[136],"Flash-kmeans":[137],"introduces":[138],"two":[139],"core":[140],"kernel-level":[141],"innovations:":[142],"(1)":[143],"FlashAssign,":[144],"which":[145,162],"fuses":[146],"computation":[148],"with":[149],"argmin":[152],"completely":[154],"bypass":[155],"intermediate":[156],"memory":[157],"materialization;":[158],"(2)":[159],"sort-inverse":[160],"update,":[161],"explicitly":[163],"constructs":[164],"inverse":[166],"mapping":[167],"transform":[169],"high-contention":[170],"scatters":[172],"into":[173],"high-bandwidth,":[174],"segment-level":[175],"localized":[176],"reductions.":[177],"Furthermore,":[178],"integrate":[180],"algorithm-system":[181],"co-designs,":[182],"including":[183],"chunked-stream":[184],"overlap":[185],"cache-aware":[187],"compile":[188],"heuristics,":[189],"ensure":[191],"practical":[192],"deployability.":[193],"Extensive":[194],"evaluations":[195],"on":[196],"NVIDIA":[197],"H200":[198],"GPUs":[199],"demonstrate":[200],"flash-kmeans":[202],"achieves":[203],"up":[204],"17.9$\\times$":[206],"end-to-end":[207],"speedup":[208],"over":[209,223],"best":[210],"baselines,":[211],"while":[212],"outperforming":[213],"industry-standard":[214],"libraries":[215],"like":[216],"cuML":[217],"FAISS":[219],"33$\\times$":[221],"200$\\times$,":[224],"respectively.":[225],"Our":[226],"code":[227],"open-sourced":[229],"at":[230],"https://github.com/svg-project/flash-kmeans.":[231]},"counts_by_year":[],"updated_date":"2026-04-14T06:02:45.956762","created_date":"2026-03-12T00:00:00"}