{"id":"https://openalex.org/W3110271845","doi":null,"title":"EdgeBERT: Optimizing On-Chip Inference for Multi-Task NLP.","display_name":"EdgeBERT: Optimizing On-Chip Inference for Multi-Task NLP.","publication_year":2020,"publication_date":"2020-11-28","ids":{"openalex":"https://openalex.org/W3110271845","mag":"3110271845"},"language":"en","primary_location":{"id":"mag:3110271845","doi":null,"is_oa":false,"landing_page_url":"http://export.arxiv.org/pdf/2011.14203","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":null},"volume":null,"type":"article","is_traditional":true,"indexed_in":[],"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/A5005762501","display_name":"Thierry Tambe","orcid":"https://orcid.org/0000-0002-6411-9620","observed_orcids":["https://orcid.org/0000-0002-6411-9620"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Thierry Tambe","raw_affiliation_strings":[],"raw_orcid":"https://orcid.org/0000-0002-6411-9620","affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5007438902","display_name":"Coleman Richard Charles Hooper","orcid":"https://orcid.org/0000-0002-5890-610X","observed_orcids":["https://orcid.org/0000-0002-5890-610X"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Coleman Hooper","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5086251333","display_name":"Lillian Pentecost","orcid":"https://orcid.org/0000-0002-6204-1938","observed_orcids":["https://orcid.org/0000-0002-6204-1938"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lillian Pentecost","raw_affiliation_strings":[],"raw_orcid":"https://orcid.org/0000-0002-6204-1938","affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5088786790","display_name":"En-Yu Yang","orcid":"https://orcid.org/0000-0003-0281-4086","observed_orcids":["https://orcid.org/0000-0003-0281-4086"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"En-Yu Yang","raw_affiliation_strings":[],"raw_orcid":"https://orcid.org/0000-0003-0281-4086","affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102005543","display_name":"Marco Donato","orcid":"https://orcid.org/0000-0002-9354-3447","observed_orcids":["https://orcid.org/0000-0002-9354-3447"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Marco Donato","raw_affiliation_strings":[],"raw_orcid":"https://orcid.org/0000-0002-9354-3447","affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5049805631","display_name":"Victor Sanh","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Victor Sanh","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5062546146","display_name":"Alexander M. Rush","orcid":"https://orcid.org/0000-0002-9900-1606","observed_orcids":["https://orcid.org/0000-0002-9900-1606"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Alexander M. Rush","raw_affiliation_strings":[],"raw_orcid":"https://orcid.org/0000-0002-9900-1606","affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5026496503","display_name":"David M. Brooks","orcid":"https://orcid.org/0000-0002-0662-7889","observed_orcids":["https://orcid.org/0000-0002-0662-7889"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"David Brooks","raw_affiliation_strings":[],"raw_orcid":"https://orcid.org/0000-0002-0662-7889","affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5043327132","display_name":"Gu-Yeon Wei","orcid":"https://orcid.org/0000-0001-5730-9904","observed_orcids":["https://orcid.org/0000-0001-5730-9904"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gu-Yeon Wei","raw_affiliation_strings":[],"raw_orcid":"https://orcid.org/0000-0001-5730-9904","affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.672,"has_fulltext":false,"cited_by_count":8,"citation_normalized_percentile":{"value":0.66234786,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":93,"max":97},"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/T10036","display_name":"Advanced Neural Network Applications","score":0.9390000104904175,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.9390000104904175,"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/T10273","display_name":"IoT and Edge/Fog Computing","score":0.018699999898672104,"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.010300000198185444,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/bert","display_name":"BERT","score":0.7770000100135803},{"id":"https://openalex.org/keywords/edge-computing","display_name":"edge computing","score":0.6940000057220459},{"id":"https://openalex.org/keywords/inference-latency","display_name":"inference latency","score":0.7870000004768372},{"id":"https://openalex.org/keywords/energy-efficiency","display_name":"energy efficiency","score":0.8870000243186951},{"id":"https://openalex.org/keywords/pruning","display_name":"pruning","score":0.44699999690055847},{"id":"https://openalex.org/keywords/floating-point-quantization","display_name":"floating-point quantization","score":0.9860000014305115},{"id":"https://openalex.org/keywords/non-volatile-memory","display_name":"non-volatile memory","score":0.9070000052452087},{"id":"https://openalex.org/keywords/hardware-accelerators","display_name":"hardware accelerators","score":0.5789999961853027}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8246524333953857},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6691651344299316},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.4792861342430115},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.43349263072013855},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.43119657039642334},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.428713321685791},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.4269362986087799},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.41599270701408386},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.10174143314361572},{"id":"https://openalex.org/C165801399","wikidata":"https://www.wikidata.org/wiki/Q25428","display_name":"Voltage","level":2,"score":0.0},{"id":"https://openalex.org/C62520636","wikidata":"https://www.wikidata.org/wiki/Q944","display_name":"Quantum mechanics","level":1,"score":0.0},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"mag:3110271845","doi":null,"is_oa":false,"landing_page_url":"http://export.arxiv.org/pdf/2011.14203","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":null}],"best_oa_location":null,"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":51,"referenced_works":["https://openalex.org/W1966939297","https://openalex.org/W1981321271","https://openalex.org/W2000967104","https://openalex.org/W2010202670","https://openalex.org/W2020740707","https://openalex.org/W2063799733","https://openalex.org/W2152839228","https://openalex.org/W2285660444","https://openalex.org/W2513554817","https://openalex.org/W2515287984","https://openalex.org/W2516141709","https://openalex.org/W2606722458","https://openalex.org/W2625457103","https://openalex.org/W2626778328","https://openalex.org/W2626991402","https://openalex.org/W2796675485","https://openalex.org/W2809188712","https://openalex.org/W2883030312","https://openalex.org/W2883283076","https://openalex.org/W2883542588","https://openalex.org/W2883899390","https://openalex.org/W2883920103","https://openalex.org/W2945969196","https://openalex.org/W2946567085","https://openalex.org/W2949117887","https://openalex.org/W2962677625","https://openalex.org/W2963310665","https://openalex.org/W2963341956","https://openalex.org/W2964299589","https://openalex.org/W2964308564","https://openalex.org/W2965373594","https://openalex.org/W2970565456","https://openalex.org/W2975059944","https://openalex.org/W2977250065","https://openalex.org/W2978017171","https://openalex.org/W2979314664","https://openalex.org/W2980104813","https://openalex.org/W2980282514","https://openalex.org/W2980965328","https://openalex.org/W2997929983","https://openalex.org/W2998183051","https://openalex.org/W3006586535","https://openalex.org/W3007007518","https://openalex.org/W3007303655","https://openalex.org/W3034457371","https://openalex.org/W3035038672","https://openalex.org/W3035251378","https://openalex.org/W3037132819","https://openalex.org/W3042501387","https://openalex.org/W3196923642","https://openalex.org/W4242577057"],"related_works":["https://openalex.org/W3207622241","https://openalex.org/W4361229345","https://openalex.org/W4312393584","https://openalex.org/W7162335977","https://openalex.org/W7163533105","https://openalex.org/W3184454880","https://openalex.org/W7123361559","https://openalex.org/W4303645613","https://openalex.org/W4312964950","https://openalex.org/W7066156550"],"abstract_inverted_index":{"Transformer-based":[0],"language":[1,15],"models":[2],"such":[3],"as":[4],"BERT":[5],"provide":[6],"significant":[7],"accuracy":[8,86],"improvement":[9],"to":[10,29,31,51,63,70,117],"a":[11,93,133],"multitude":[12],"of":[13,96,121,141,160,171],"natural":[14],"processing":[16],"(NLP)":[17],"tasks.":[18],"However,":[19],"their":[20],"hefty":[21],"computational":[22],"and":[23,44,47,55,72,76,107,110,126,165,175],"memory":[24,77],"demands":[25],"make":[26],"them":[27],"challenging":[28],"deploy":[30],"resource-constrained":[32],"edge":[33,128],"platforms":[34],"with":[35,80,163],"strict":[36],"latency":[37,54,75],"requirements.":[38],"\r\nWe":[39],"present":[40],"EdgeBERT":[41,154],"an":[42,172,179],"in-depth":[43],"principled":[45],"algorithm":[46],"hardware":[48,135],"design":[49],"methodology":[50],"achieve":[52,68],"minimal":[53],"energy":[56,168],"consumption":[57],"on":[58,87,178],"multi-task":[59,144],"NLP":[60,161],"inference.":[61],"Compared":[62],"the":[64,119,142],"ALBERT":[65],"baseline,":[66],"we":[67,131],"up":[69],"2.4x":[71],"13.4x":[73],"inference":[74,158],"savings,":[78],"respectively,":[79],"less":[81],"than":[82,169],"1%-pt":[83],"drop":[84],"in":[85,115,124,149],"several":[88],"GLUE":[89],"benchmarks":[90],"by":[91],"employing":[92],"calibrated":[94],"combination":[95],"1)":[97],"entropy-based":[98],"early":[99],"stopping,":[100],"2)":[101],"adaptive":[102],"attention":[103],"span,":[104],"3)":[105],"movement":[106],"magnitude":[108],"pruning,":[109],"4)":[111],"floating-point":[112,138],"quantization.":[113],"\r\nFurthermore,":[114],"order":[116],"maximize":[118],"benefits":[120],"these":[122],"algorithms":[123],"always-on":[125],"intermediate":[127],"computing":[129],"settings,":[130],"specialize":[132],"scalable":[134],"architecture":[136],"wherein":[137],"bit":[139],"encodings":[140],"shareable":[143],"embedding":[145],"parameters":[146],"are":[147],"stored":[148],"high-density":[150],"non-volatile":[151],"memory.":[152],"Altogether,":[153],"enables":[155],"fully":[156],"on-chip":[157],"acceleration":[159],"workloads":[162],"5.2x,":[164],"157x":[166],"lower":[167],"that":[170],"un-optimized":[173],"accelerator":[174],"CUDA":[176],"adaptations":[177],"Nvidia":[180],"Jetson":[181],"Tegra":[182],"X2":[183],"mobile":[184],"GPU,":[185],"respectively.":[186]},"counts_by_year":[{"year":2023,"cited_by_count":2},{"year":2022,"cited_by_count":2},{"year":2021,"cited_by_count":4}],"updated_date":"2026-10-16T09:59:22.235173","created_date":"2020-12-07T00:00:00"}