{"id":"https://openalex.org/W4290943581","doi":"https://doi.org/10.1145/3534678.3539132","title":"BE3R: BERT based Early-Exit Using Expert Routing","display_name":"BE3R: BERT based Early-Exit Using Expert Routing","publication_year":2022,"publication_date":"2022-08-12","ids":{"openalex":"https://openalex.org/W4290943581","doi":"https://doi.org/10.1145/3534678.3539132"},"language":"en","primary_location":{"id":"doi:10.1145/3534678.3539132","is_oa":false,"landing_page_url":"https://doi.org/10.1145/3534678.3539132","pdf_url":null,"source":{"id":"https://openalex.org/S4363608767","display_name":"Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining","issn_l":null,"issn":null,"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":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining","raw_type":"proceedings-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/A5048579807","display_name":"Sourab Mangrulkar","orcid":"https://orcid.org/0000-0002-7997-1170"},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Sourab Mangrulkar","raw_affiliation_strings":["Amazon, Bengaluru, India"],"affiliations":[{"raw_affiliation_string":"Amazon, Bengaluru, India","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5087399856","display_name":"M S Ankith","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ankith M S","raw_affiliation_strings":["Amazon, Bengaluru, India"],"affiliations":[{"raw_affiliation_string":"Amazon, Bengaluru, India","institution_ids":[]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5038625048","display_name":"Vivek Sembium","orcid":"https://orcid.org/0000-0002-6787-1383"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Vivek Sembium","raw_affiliation_strings":["Amazon, Bengaluru, India"],"affiliations":[{"raw_affiliation_string":"Amazon, Bengaluru, India","institution_ids":[]}]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":3,"corresponding_author_ids":["https://openalex.org/A5048579807"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.3118,"has_fulltext":false,"cited_by_count":3,"citation_normalized_percentile":{"value":0.4984917,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":89,"max":96},"biblio":{"volume":null,"issue":null,"first_page":"3504","last_page":"3512"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.9997000098228455,"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"}},"topics":[{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.9997000098228455,"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"}},{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.9975000023841858,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.9970999956130981,"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/computer-science","display_name":"Computer science","score":0.8103581666946411},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.7316370010375977},{"id":"https://openalex.org/keywords/benchmark","display_name":"Benchmark (surveying)","score":0.6867523789405823},{"id":"https://openalex.org/keywords/routing","display_name":"Routing (electronic design automation)","score":0.618364155292511},{"id":"https://openalex.org/keywords/metric","display_name":"Metric (unit)","score":0.5742318034172058},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.5624902844429016},{"id":"https://openalex.org/keywords/relevance","display_name":"Relevance (law)","score":0.5258025527000427},{"id":"https://openalex.org/keywords/machine-learning","display_name":"Machine learning","score":0.502424955368042},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.4864611327648163},{"id":"https://openalex.org/keywords/throughput","display_name":"Throughput","score":0.4751577079296112},{"id":"https://openalex.org/keywords/computer-network","display_name":"Computer network","score":0.12213176488876343},{"id":"https://openalex.org/keywords/operating-system","display_name":"Operating system","score":0.09194058179855347}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8103581666946411},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.7316370010375977},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.6867523789405823},{"id":"https://openalex.org/C74172769","wikidata":"https://www.wikidata.org/wiki/Q1446839","display_name":"Routing (electronic design automation)","level":2,"score":0.618364155292511},{"id":"https://openalex.org/C176217482","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Metric (unit)","level":2,"score":0.5742318034172058},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.5624902844429016},{"id":"https://openalex.org/C158154518","wikidata":"https://www.wikidata.org/wiki/Q7310970","display_name":"Relevance (law)","level":2,"score":0.5258025527000427},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.502424955368042},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4864611327648163},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.4751577079296112},{"id":"https://openalex.org/C31258907","wikidata":"https://www.wikidata.org/wiki/Q1301371","display_name":"Computer network","level":1,"score":0.12213176488876343},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.09194058179855347},{"id":"https://openalex.org/C17744445","wikidata":"https://www.wikidata.org/wiki/Q36442","display_name":"Political science","level":0,"score":0.0},{"id":"https://openalex.org/C205649164","wikidata":"https://www.wikidata.org/wiki/Q1071","display_name":"Geography","level":0,"score":0.0},{"id":"https://openalex.org/C21547014","wikidata":"https://www.wikidata.org/wiki/Q1423657","display_name":"Operations management","level":1,"score":0.0},{"id":"https://openalex.org/C555944384","wikidata":"https://www.wikidata.org/wiki/Q249","display_name":"Wireless","level":2,"score":0.0},{"id":"https://openalex.org/C199539241","wikidata":"https://www.wikidata.org/wiki/Q7748","display_name":"Law","level":1,"score":0.0},{"id":"https://openalex.org/C13280743","wikidata":"https://www.wikidata.org/wiki/Q131089","display_name":"Geodesy","level":1,"score":0.0},{"id":"https://openalex.org/C162324750","wikidata":"https://www.wikidata.org/wiki/Q8134","display_name":"Economics","level":0,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3534678.3539132","is_oa":false,"landing_page_url":"https://doi.org/10.1145/3534678.3539132","pdf_url":null,"source":{"id":"https://openalex.org/S4363608767","display_name":"Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining","issn_l":null,"issn":null,"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":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":28,"referenced_works":["https://openalex.org/W1516184288","https://openalex.org/W1821462560","https://openalex.org/W2004298296","https://openalex.org/W2750779823","https://openalex.org/W2923014074","https://openalex.org/W2963748441","https://openalex.org/W2963846996","https://openalex.org/W2970641574","https://openalex.org/W2978017171","https://openalex.org/W2978670439","https://openalex.org/W3008374555","https://openalex.org/W3033188311","https://openalex.org/W3033529678","https://openalex.org/W3034439313","https://openalex.org/W3034457371","https://openalex.org/W3035030897","https://openalex.org/W3035038672","https://openalex.org/W3101163004","https://openalex.org/W3101248447","https://openalex.org/W3105966348","https://openalex.org/W4287391717","https://openalex.org/W4287704453","https://openalex.org/W4287901267","https://openalex.org/W4295838474","https://openalex.org/W4309793872","https://openalex.org/W4323654151","https://openalex.org/W4391156274","https://openalex.org/W6636625392"],"related_works":["https://openalex.org/W2136583354","https://openalex.org/W2111238207","https://openalex.org/W2055243143","https://openalex.org/W2760721665","https://openalex.org/W2112044895","https://openalex.org/W330130819","https://openalex.org/W2288610023","https://openalex.org/W3121416282","https://openalex.org/W2281389338","https://openalex.org/W2037453743"],"abstract_inverted_index":{"Pre-trained":[0],"language":[1],"models":[2,165],"like":[3],"BERT":[4],"have":[5],"reported":[6],"state-of-the-art":[7],"performance":[8,135,161],"on":[9,123,133,154],"several":[10],"Natural":[11],"Language":[12],"Processing":[13],"(NLP)":[14],"tasks,":[15],"but":[16],"high":[17],"computational":[18],"demands":[19],"hinder":[20],"its":[21],"widespread":[22],"adoption":[23],"for":[24],"large":[25,97,124],"scale":[26],"NLP":[27],"tasks.":[28],"In":[29],"this":[30],"work,":[31],"we":[32,50],"propose":[33],"a":[34,78],"novel":[35],"routing":[36],"based":[37,44],"early":[38],"exit":[39,54],"model":[40,116,122],"called":[41],"BE3R":[42,105],"(BERT":[43],"Early-Exit":[45],"using":[46,141],"Expert":[47],"Routing),":[48],"where":[49],"learn":[51],"to":[52,61,77,162],"dynamically":[53],"in":[55,91,173],"the":[56,64,68,84,107,114,120,134],"earlier":[57],"layers":[58],"without":[59,130],"needing":[60],"traverse":[62],"through":[63],"entire":[65],"model.":[66],"Unlike":[67],"exiting":[69],"early-exit":[70],"methods,":[71],"our":[72],"approach":[73],"can":[74],"be":[75],"extended":[76],"batch":[79,108,174],"inference":[80,109,175],"setting.":[81,176],"We":[82,137,151],"consider":[83],"specific":[85],"application":[86],"of":[87],"search":[88],"relevance":[89],"filtering":[90],"Amazon":[92],"India":[93],"marketplace":[94],"services":[95],"(a":[96],"e-commerce":[98],"website).":[99],"Our":[100],"experimental":[101],"results":[102],"show":[103],"that":[104],"improves":[106],"throughput":[110,169],"by":[111],"46.5%":[112],"over":[113,119],"BERT-Base":[115],"and":[117,147,158],"35.89%":[118],"DistilBERT-Base":[121],"dataset":[125],"with":[126,166],"50":[127],"Million":[128],"samples":[129],"any":[131],"trade-off":[132],"metric.":[136],"conduct":[138],"thorough":[139],"experimentation":[140],"various":[142],"architectural":[143],"choices,":[144],"loss":[145],"functions":[146],"perform":[148,152],"qualitative":[149],"analysis.":[150],"experiments":[153],"public":[155],"GLUE":[156],"Benchmark":[157],"demonstrate":[159],"comparable":[160],"corresponding":[163],"baseline":[164],"23%":[167],"average":[168],"improvement":[170],"across":[171],"tasks":[172]},"counts_by_year":[{"year":2025,"cited_by_count":2},{"year":2023,"cited_by_count":1}],"updated_date":"2025-11-06T03:46:38.306776","created_date":"2025-10-10T00:00:00"}