{"id":"https://openalex.org/W4389162698","doi":"https://doi.org/10.1109/iccad57390.2023.10323651","title":"Edge-MoE: Memory-Efficient Multi-Task Vision Transformer Architecture with Task-Level Sparsity via Mixture-of-Experts","display_name":"Edge-MoE: Memory-Efficient Multi-Task Vision Transformer Architecture with Task-Level Sparsity via Mixture-of-Experts","publication_year":2023,"publication_date":"2023-10-28","ids":{"openalex":"https://openalex.org/W4389162698","doi":"https://doi.org/10.1109/iccad57390.2023.10323651"},"language":"en","primary_location":{"id":"doi:10.1109/iccad57390.2023.10323651","is_oa":false,"landing_page_url":"https://doi.org/10.1109/iccad57390.2023.10323651","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2023 IEEE/ACM International Conference on Computer Aided Design (ICCAD)","raw_type":"proceedings-article"},"type":"conference-paper","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/A5032302447","display_name":"Rishov Sarkar","orcid":"https://orcid.org/0000-0002-9168-0392"},"institutions":[{"id":"https://openalex.org/I130701444","display_name":"Georgia Institute of Technology","ror":"https://ror.org/01zkghx44","country_code":"US","type":"education","lineage":["https://openalex.org/I130701444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Rishov Sarkar","raw_affiliation_strings":["School of Electrical and Computer Engineering, Georgia Institute of Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Electrical and Computer Engineering, Georgia Institute of Technology","institution_ids":["https://openalex.org/I130701444"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5114104110","display_name":"Hanxue Liang","orcid":null},"institutions":[{"id":"https://openalex.org/I86519309","display_name":"The University of Texas at Austin","ror":"https://ror.org/00hj54h04","country_code":"US","type":"education","lineage":["https://openalex.org/I86519309"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Hanxue Liang","raw_affiliation_strings":["School of Electrical and Computer Engineering, University of Texas at Austin"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Electrical and Computer Engineering, University of Texas at Austin","institution_ids":["https://openalex.org/I86519309"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5067582681","display_name":"Zhiwen Fan","orcid":null},"institutions":[{"id":"https://openalex.org/I86519309","display_name":"The University of Texas at Austin","ror":"https://ror.org/00hj54h04","country_code":"US","type":"education","lineage":["https://openalex.org/I86519309"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zhiwen Fan","raw_affiliation_strings":["School of Electrical and Computer Engineering, University of Texas at Austin"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Electrical and Computer Engineering, University of Texas at Austin","institution_ids":["https://openalex.org/I86519309"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5048522863","display_name":"Zhangyang Wang","orcid":"https://orcid.org/0000-0002-2050-5693"},"institutions":[{"id":"https://openalex.org/I86519309","display_name":"The University of Texas at Austin","ror":"https://ror.org/00hj54h04","country_code":"US","type":"education","lineage":["https://openalex.org/I86519309"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zhangyang Wang","raw_affiliation_strings":["School of Electrical and Computer Engineering, University of Texas at Austin"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Electrical and Computer Engineering, University of Texas at Austin","institution_ids":["https://openalex.org/I86519309"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5053022655","display_name":"Cong Hao","orcid":"https://orcid.org/0000-0002-2541-8767"},"institutions":[{"id":"https://openalex.org/I130701444","display_name":"Georgia Institute of Technology","ror":"https://ror.org/01zkghx44","country_code":"US","type":"education","lineage":["https://openalex.org/I130701444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Cong Hao","raw_affiliation_strings":["School of Electrical and Computer Engineering, Georgia Institute of Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Electrical and Computer Engineering, Georgia Institute of Technology","institution_ids":["https://openalex.org/I130701444"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":25.4206,"has_fulltext":false,"cited_by_count":47,"citation_normalized_percentile":{"value":0.99879719,"is_in_top_1_percent":true,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":98,"max":100},"biblio":{"volume":null,"issue":null,"first_page":"01","last_page":"09"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11992","display_name":"CCD and CMOS Imaging Sensors","score":0.9998999834060669,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T11992","display_name":"CCD and CMOS Imaging Sensors","score":0.9998999834060669,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10502","display_name":"Advanced Memory and Neural Computing","score":0.9993000030517578,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.9980000257492065,"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/computer-science","display_name":"Computer science","score":0.7068368196487427},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.6724909543991089},{"id":"https://openalex.org/keywords/architecture","display_name":"Architecture","score":0.6603886485099792},{"id":"https://openalex.org/keywords/task","display_name":"Task (project management)","score":0.6377471089363098},{"id":"https://openalex.org/keywords/computer-architecture","display_name":"Computer architecture","score":0.4398650527000427},{"id":"https://openalex.org/keywords/task-analysis","display_name":"Task analysis","score":0.43497970700263977},{"id":"https://openalex.org/keywords/human\u2013computer-interaction","display_name":"Human\u2013computer interaction","score":0.41342535614967346},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.3933488428592682},{"id":"https://openalex.org/keywords/embedded-system","display_name":"Embedded system","score":0.3524409532546997},{"id":"https://openalex.org/keywords/engineering","display_name":"Engineering","score":0.16111135482788086},{"id":"https://openalex.org/keywords/electrical-engineering","display_name":"Electrical engineering","score":0.09954845905303955},{"id":"https://openalex.org/keywords/systems-engineering","display_name":"Systems engineering","score":0.0938495397567749},{"id":"https://openalex.org/keywords/voltage","display_name":"Voltage","score":0.08926078677177429}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7068368196487427},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.6724909543991089},{"id":"https://openalex.org/C123657996","wikidata":"https://www.wikidata.org/wiki/Q12271","display_name":"Architecture","level":2,"score":0.6603886485099792},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.6377471089363098},{"id":"https://openalex.org/C118524514","wikidata":"https://www.wikidata.org/wiki/Q173212","display_name":"Computer architecture","level":1,"score":0.4398650527000427},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.43497970700263977},{"id":"https://openalex.org/C107457646","wikidata":"https://www.wikidata.org/wiki/Q207434","display_name":"Human\u2013computer interaction","level":1,"score":0.41342535614967346},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3933488428592682},{"id":"https://openalex.org/C149635348","wikidata":"https://www.wikidata.org/wiki/Q193040","display_name":"Embedded system","level":1,"score":0.3524409532546997},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.16111135482788086},{"id":"https://openalex.org/C119599485","wikidata":"https://www.wikidata.org/wiki/Q43035","display_name":"Electrical engineering","level":1,"score":0.09954845905303955},{"id":"https://openalex.org/C201995342","wikidata":"https://www.wikidata.org/wiki/Q682496","display_name":"Systems engineering","level":1,"score":0.0938495397567749},{"id":"https://openalex.org/C165801399","wikidata":"https://www.wikidata.org/wiki/Q25428","display_name":"Voltage","level":2,"score":0.08926078677177429},{"id":"https://openalex.org/C153349607","wikidata":"https://www.wikidata.org/wiki/Q36649","display_name":"Visual arts","level":1,"score":0.0},{"id":"https://openalex.org/C142362112","wikidata":"https://www.wikidata.org/wiki/Q735","display_name":"Art","level":0,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/iccad57390.2023.10323651","is_oa":false,"landing_page_url":"https://doi.org/10.1109/iccad57390.2023.10323651","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2023 IEEE/ACM International Conference on Computer Aided Design (ICCAD)","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":40,"referenced_works":["https://openalex.org/W2340897893","https://openalex.org/W2798441115","https://openalex.org/W2809290718","https://openalex.org/W2895401575","https://openalex.org/W2896457183","https://openalex.org/W2899663614","https://openalex.org/W2959581809","https://openalex.org/W2963430933","https://openalex.org/W2963877604","https://openalex.org/W2964185501","https://openalex.org/W2964247799","https://openalex.org/W2966182616","https://openalex.org/W3047848469","https://openalex.org/W3085046840","https://openalex.org/W3094502228","https://openalex.org/W3096609285","https://openalex.org/W3097571420","https://openalex.org/W3121523901","https://openalex.org/W3133696297","https://openalex.org/W3138516171","https://openalex.org/W3138895808","https://openalex.org/W3157506437","https://openalex.org/W3162542754","https://openalex.org/W3171125843","https://openalex.org/W3176468986","https://openalex.org/W3196923642","https://openalex.org/W3199934250","https://openalex.org/W3211983116","https://openalex.org/W3214728273","https://openalex.org/W4214612132","https://openalex.org/W4226549668","https://openalex.org/W4295856455","https://openalex.org/W4312910119","https://openalex.org/W4321637298","https://openalex.org/W6755207826","https://openalex.org/W6755977528","https://openalex.org/W6788135285","https://openalex.org/W6790690058","https://openalex.org/W6795140394","https://openalex.org/W6845510015"],"related_works":["https://openalex.org/W2588198209","https://openalex.org/W1909006023","https://openalex.org/W4205824991","https://openalex.org/W3200723557","https://openalex.org/W4312713546","https://openalex.org/W2362195430","https://openalex.org/W2347494122","https://openalex.org/W2567983276","https://openalex.org/W2802298219","https://openalex.org/W2038503502"],"abstract_inverted_index":{"The":[0],"computer":[1],"vision":[2],"community":[3],"is":[4,60,86,159,398],"embracing":[5],"two":[6],"promising":[7,198],"learning":[8],"paradigms:":[9],"the":[10,34,68,87,110,123,132,157,173,188,194,197,230,239,270,279,299,329],"Vision":[11],"Transformer":[12],"(ViT)":[13],"and":[14,105,116,121,138,164,172,210,229,291,302,305,309,354,363,370,379,414],"Multi-task":[15],"Learning":[16],"(MTL).":[17],"ViT":[18,246,392],"models":[19,228],"show":[20],"extraordinary":[21],"performance":[22,49,416],"over":[23,117],"traditional":[24],"convolution":[25],"networks":[26],"but":[27,56],"are":[28,79,103,142,177],"commonly":[29],"recognized":[30],"as":[31,161],"computation-intensive,":[32],"especially":[33],"self-attention":[35,158],"with":[36,47,234,247,333,360,376,403],"quadratic":[37],"complexity.":[38],"MTL":[39,63,128],"uses":[40],"one":[41,74],"model":[42,70],"to":[43,275,322,342],"infer":[44],"multiple":[45],"tasks":[46,78],"better":[48,114,372],"by":[50,317],"enforcing":[51],"shared":[52,316],"representation":[53],"among":[54],"tasks,":[55],"a":[57,76,97,248,262,283,307,337],"huge":[58],"drawback":[59],"that,":[61],"most":[62],"regimes":[64],"require":[65],"activation":[66,174],"of":[67,100,135,187,216,251,278,388],"entire":[69,189],"even":[71,139],"when":[72],"only":[73,96],"or":[75],"few":[77],"needed,":[80],"causing":[81],"significant":[82],"computing":[83,311],"waste.":[84],"M3ViT":[85,195,233,332],"latest":[88],"multi-task":[89,202,232,245,331,391],"Vi$T$model":[90],"that":[91,313],"introduces":[92],"mixture-of-experts":[93],"(MoE),":[94],"where":[95],"small":[98],"portion":[99],"subnetworks":[101],"(\u201cexperts\u201d)":[102],"sparsely":[104],"dynamically":[106],"activated":[107],"based":[108],"on":[109,149,356,393],"current":[111],"task.":[112],"M3Vi$T$achieves":[113],"accuracy":[115],"80%":[118],"computation":[119,300],"reduction":[120],"paves":[122],"way":[124],"for":[125,146,201,206,244,255,266,328],"efficient":[126,147],"real-time":[127,385],"using":[129,407],"ViT.":[130],"Despite":[131],"algorithmic":[133],"advantages":[134],"MTL,":[136],"ViT,":[137],"M3ViT,":[140],"there":[141],"still":[143],"many":[144],"challenges":[145,205,223],"deployment":[148],"FPGA.":[150],"For":[151],"instance,":[152],"in":[153,224,400],"general":[154,226,256],"Transformer/ViT":[155,227,257],"models,":[156,258],"known":[160],"computational":[162,320],"intensive":[163],"requires":[165],"high":[166],"bandwidth.":[167],"In":[168,193],"addition,":[169],"softmax":[170,286],"operations":[171],"function":[175],"GELU":[176,293],"extensively":[178],"used,":[179],"which":[180,268,295,367],"unfortunately":[181],"can":[182,296,314],"consume":[183],"more":[184,217],"than":[185],"half":[186],"FPGA":[190,242,405,412],"resource":[191,213,303,325],"(LUTs).":[192],"model,":[196],"MoE":[199],"mechanism":[200,265],"exposes":[203],"new":[204],"memory":[207,346],"access":[208,347],"overhead":[209],"also":[211],"increases":[212],"usage":[214],"because":[215],"layer":[218],"types.":[219],"To":[220],"address":[221],"these":[222],"both":[225],"state-of-the-art":[231],"MoE,":[235,334],"we":[236,259,335,350],"propose":[237,260,336],"Edge-MoE,":[238],"first":[240],"end-to-end":[241],"accelerator":[243],"rich":[249],"collection":[250],"architectural":[252],"innovations.":[253],"First,":[254],"(1)":[261],"novel":[263,338],"reordering":[264,340],"self-attention,":[267],"reduces":[269],"bandwidth":[271],"requirement":[272],"from":[273],"proportional":[274],"constant":[276],"regardless":[277],"target":[280],"parallelism;":[281],"(2)":[282],"fast":[284],"single-pass":[285],"approximation;":[287],"(3)":[288],"an":[289,394],"accurate":[290],"low-cost":[292],"approximation,":[294],"significantly":[297],"reduce":[298,324],"latency":[301],"usage;":[304],"(4)":[306],"unified":[308],"flexible":[310],"unit":[312],"be":[315],"almost":[318],"all":[319],"layers":[321],"maximally":[323],"usage.":[326],"Second,":[327],"advanced":[330],"patch":[339],"method":[341],"completely":[343],"eliminate":[344],"any":[345],"overhead.":[348],"Third,":[349],"deliver":[351],"on-board":[352,415],"implementation":[353],"measurement":[355],"Xilinx":[357],"ZCU102":[358],"FPGA,":[359],"verified":[361],"functionality":[362],"open-sourced":[364],"hardware":[365],"design,":[366],"achieves":[368],"2.24\u00d7":[369],"4.90\u00d7":[371],"energy":[373],"efficiency":[374],"comparing":[375],"GPU":[377],"(A6000)":[378],"CPU":[380],"(Xeon":[381],"6226R),":[382],"respectively.":[383],"A":[384],"video":[386],"demonstration":[387],"our":[389,404],"accelerated":[390],"autonomous":[395],"driving":[396],"dataset":[397],"available":[399],"GitHub,11https://github.com/sharc-lab/Edge-MoE/raw/main/demo.mp4":[401],"together":[402],"design":[406],"High-Level":[408],"Synthesis,":[409],"host":[410],"code,":[411],"bitstream,":[413],"results.":[417]},"counts_by_year":[{"year":2026,"cited_by_count":16},{"year":2025,"cited_by_count":25},{"year":2024,"cited_by_count":6}],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2025-10-10T00:00:00"}