{"id":"https://openalex.org/W3015662800","doi":"https://doi.org/10.1109/icassp40776.2020.9054324","title":"How Much Self-Attention Do We Need\u0192 Trading Attention for Feed-Forward Layers","display_name":"How Much Self-Attention Do We Need\u0192 Trading Attention for Feed-Forward Layers","publication_year":2020,"publication_date":"2020-04-09","ids":{"openalex":"https://openalex.org/W3015662800","doi":"https://doi.org/10.1109/icassp40776.2020.9054324","mag":"3015662800"},"language":"en","primary_location":{"id":"doi:10.1109/icassp40776.2020.9054324","is_oa":false,"landing_page_url":"https://doi.org/10.1109/icassp40776.2020.9054324","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)","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/A5002810304","display_name":"Kazuki Irie","orcid":null},"institutions":[{"id":"https://openalex.org/I887968799","display_name":"RWTH Aachen University","ror":"https://ror.org/04xfq0f34","country_code":"DE","type":"education","lineage":["https://openalex.org/I887968799"]}],"countries":["DE"],"is_corresponding":true,"raw_author_name":"Kazuki Irie","raw_affiliation_strings":["Human Language Technology and Pattern Recognition Group, RWTH Aachen University, Aachen, Germany"],"affiliations":[{"raw_affiliation_string":"Human Language Technology and Pattern Recognition Group, RWTH Aachen University, Aachen, Germany","institution_ids":["https://openalex.org/I887968799"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5010929553","display_name":"Alexander Gerstenberger","orcid":null},"institutions":[{"id":"https://openalex.org/I887968799","display_name":"RWTH Aachen University","ror":"https://ror.org/04xfq0f34","country_code":"DE","type":"education","lineage":["https://openalex.org/I887968799"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Alexander Gerstenberger","raw_affiliation_strings":["Human Language Technology and Pattern Recognition Group, RWTH Aachen University, Aachen, Germany"],"affiliations":[{"raw_affiliation_string":"Human Language Technology and Pattern Recognition Group, RWTH Aachen University, Aachen, Germany","institution_ids":["https://openalex.org/I887968799"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5088968292","display_name":"Ralf Schl\u00fcter","orcid":"https://orcid.org/0000-0003-2839-9247"},"institutions":[{"id":"https://openalex.org/I887968799","display_name":"RWTH Aachen University","ror":"https://ror.org/04xfq0f34","country_code":"DE","type":"education","lineage":["https://openalex.org/I887968799"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Ralf Schluter","raw_affiliation_strings":["Human Language Technology and Pattern Recognition Group, RWTH Aachen University, Aachen, Germany"],"affiliations":[{"raw_affiliation_string":"Human Language Technology and Pattern Recognition Group, RWTH Aachen University, Aachen, Germany","institution_ids":["https://openalex.org/I887968799"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5112501010","display_name":"Hermann Ney","orcid":null},"institutions":[{"id":"https://openalex.org/I887968799","display_name":"RWTH Aachen University","ror":"https://ror.org/04xfq0f34","country_code":"DE","type":"education","lineage":["https://openalex.org/I887968799"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Hermann Ney","raw_affiliation_strings":["Human Language Technology and Pattern Recognition Group, RWTH Aachen University, Aachen, Germany"],"affiliations":[{"raw_affiliation_string":"Human Language Technology and Pattern Recognition Group, RWTH Aachen University, Aachen, Germany","institution_ids":["https://openalex.org/I887968799"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":4,"corresponding_author_ids":["https://openalex.org/A5002810304"],"corresponding_institution_ids":["https://openalex.org/I887968799"],"apc_list":null,"apc_paid":null,"fwci":0.1326,"has_fulltext":false,"cited_by_count":2,"citation_normalized_percentile":{"value":0.51667575,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":89,"max":94},"biblio":{"volume":"45","issue":null,"first_page":"6154","last_page":"6158"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10201","display_name":"Speech Recognition and Synthesis","score":0.9994000196456909,"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/T10201","display_name":"Speech Recognition and Synthesis","score":0.9994000196456909,"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/T11309","display_name":"Music and Audio Processing","score":0.9961000084877014,"subfield":{"id":"https://openalex.org/subfields/1711","display_name":"Signal Processing"},"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/T10028","display_name":"Topic Modeling","score":0.9955999851226807,"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":0.6905992031097412},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.6622893810272217},{"id":"https://openalex.org/keywords/perplexity","display_name":"Perplexity","score":0.6431417465209961},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.6179294586181641},{"id":"https://openalex.org/keywords/decoding-methods","display_name":"Decoding methods","score":0.5239836573600769},{"id":"https://openalex.org/keywords/scaling","display_name":"Scaling","score":0.4792913794517517},{"id":"https://openalex.org/keywords/algorithm","display_name":"Algorithm","score":0.36613014340400696},{"id":"https://openalex.org/keywords/computer-engineering","display_name":"Computer engineering","score":0.3221641480922699},{"id":"https://openalex.org/keywords/speech-recognition","display_name":"Speech recognition","score":0.3104056715965271},{"id":"https://openalex.org/keywords/electrical-engineering","display_name":"Electrical engineering","score":0.17495641112327576},{"id":"https://openalex.org/keywords/mathematics","display_name":"Mathematics","score":0.16089269518852234},{"id":"https://openalex.org/keywords/voltage","display_name":"Voltage","score":0.12306714057922363}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6905992031097412},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.6622893810272217},{"id":"https://openalex.org/C100279451","wikidata":"https://www.wikidata.org/wiki/Q372193","display_name":"Perplexity","level":3,"score":0.6431417465209961},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.6179294586181641},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.5239836573600769},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.4792913794517517},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.36613014340400696},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.3221641480922699},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.3104056715965271},{"id":"https://openalex.org/C119599485","wikidata":"https://www.wikidata.org/wiki/Q43035","display_name":"Electrical engineering","level":1,"score":0.17495641112327576},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.16089269518852234},{"id":"https://openalex.org/C165801399","wikidata":"https://www.wikidata.org/wiki/Q25428","display_name":"Voltage","level":2,"score":0.12306714057922363},{"id":"https://openalex.org/C2524010","wikidata":"https://www.wikidata.org/wiki/Q8087","display_name":"Geometry","level":1,"score":0.0},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/icassp40776.2020.9054324","is_oa":false,"landing_page_url":"https://doi.org/10.1109/icassp40776.2020.9054324","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)","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":51,"referenced_works":["https://openalex.org/W62305045","https://openalex.org/W811578723","https://openalex.org/W1494198834","https://openalex.org/W1553004968","https://openalex.org/W1821462560","https://openalex.org/W1915251500","https://openalex.org/W1934041838","https://openalex.org/W2134797427","https://openalex.org/W2152790380","https://openalex.org/W2158195707","https://openalex.org/W2251321385","https://openalex.org/W2251640092","https://openalex.org/W2251682575","https://openalex.org/W2302255633","https://openalex.org/W2398887489","https://openalex.org/W2402144811","https://openalex.org/W2404974730","https://openalex.org/W2408727897","https://openalex.org/W2798761464","https://openalex.org/W2799923439","https://openalex.org/W2890012642","https://openalex.org/W2943845043","https://openalex.org/W2963045354","https://openalex.org/W2963088785","https://openalex.org/W2963341956","https://openalex.org/W2963403868","https://openalex.org/W2963631907","https://openalex.org/W2963660642","https://openalex.org/W2964045208","https://openalex.org/W2964110616","https://openalex.org/W2970401203","https://openalex.org/W2973215447","https://openalex.org/W2994673210","https://openalex.org/W2995575179","https://openalex.org/W2996428491","https://openalex.org/W3007776460","https://openalex.org/W3015726069","https://openalex.org/W3105532142","https://openalex.org/W4288024261","https://openalex.org/W4295838474","https://openalex.org/W6638523607","https://openalex.org/W6682948231","https://openalex.org/W6691770337","https://openalex.org/W6748148878","https://openalex.org/W6754905691","https://openalex.org/W6755207826","https://openalex.org/W6766222867","https://openalex.org/W6768021236","https://openalex.org/W6770251742","https://openalex.org/W6771626834","https://openalex.org/W6780226713"],"related_works":["https://openalex.org/W2169518243","https://openalex.org/W2252095989","https://openalex.org/W4322096525","https://openalex.org/W2551914602","https://openalex.org/W4281893144","https://openalex.org/W2105076537","https://openalex.org/W2787311093","https://openalex.org/W3013624417","https://openalex.org/W4287826556","https://openalex.org/W3049463507"],"abstract_inverted_index":{"We":[0],"propose":[1],"simple":[2],"architectural":[3],"modifications":[4],"in":[5,58,80,110,141,165],"the":[6,10,20,26,29,73,96,101,106,111,120,130,138,157,171],"standard":[7,158],"Transformer":[8,42,112,174],"with":[9,160],"goal":[11],"to":[12,49,63,93],"reduce":[13],"its":[14],"total":[15,121],"state":[16,102,139,151],"size":[17,140,152],"(defined":[18],"as":[19],"number":[21,122],"of":[22,28,38,123,132,150,167,173],"self-attention":[23],"layers":[24],"times":[25,34,154],"sum":[27],"key":[30],"and":[31,118],"value":[32],"dimensions,":[33],"position)":[35],"without":[36,99],"loss":[37,164],"performance.":[39,53],"Large":[40],"scale":[41],"language":[43,175],"models":[44,176],"have":[45],"been":[46],"empirically":[47],"proved":[48],"give":[50],"very":[51],"good":[52],"However,":[54],"scaling":[55],"up":[56],"results":[57],"a":[59,115,148],"model":[60,97,149],"that":[61],"needs":[62],"store":[64],"large":[65],"states":[66],"at":[67],"evaluation":[68],"time.":[69],"This":[70],"can":[71],"increase":[72,95],"memory":[74],"requirement":[75],"dramatically":[76],"for":[77],"search":[78],"e.g.,":[79],"speech":[81],"recognition":[82],"(first":[83],"pass":[84],"decoding,":[85],"lattice":[86],"rescoring,":[87],"or":[88],"shallow":[89],"fusion).":[90],"In":[91,125],"order":[92],"efficiently":[94],"capacity":[98],"increasing":[100],"size,":[103],"we":[104,127,146],"replace":[105],"single-layer":[107],"feed-forward":[108],"module":[109],"layer":[113],"by":[114],"deeper":[116],"network,":[117],"decrease":[119],"layers.":[124],"addition,":[126],"also":[128],"evaluate":[129],"effect":[131],"key-value":[133],"tying":[134],"which":[135,169],"directly":[136],"divides":[137],"half.":[142],"On":[143],"TED-LIUM":[144],"2,":[145],"obtain":[147],"4":[153],"smaller":[155],"than":[156],"Transformer,":[159],"only":[161],"2%":[162],"relative":[163],"terms":[166],"perplexity,":[168],"makes":[170],"deployment":[172],"more":[177],"convenient.":[178]},"counts_by_year":[{"year":2024,"cited_by_count":1},{"year":2021,"cited_by_count":1}],"updated_date":"2025-11-06T03:46:38.306776","created_date":"2025-10-10T00:00:00"}