{"id":"https://openalex.org/W3135593154","doi":"https://doi.org/10.48550/arxiv.2103.03404","title":"Attention is Not All You Need: Pure Attention Loses Rank Doubly Exponentially with Depth","display_name":"Attention is Not All You Need: Pure Attention Loses Rank Doubly Exponentially with Depth","publication_year":2021,"publication_date":"2021-03-05","ids":{"openalex":"https://openalex.org/W3135593154","doi":"https://doi.org/10.48550/arxiv.2103.03404","mag":"3135593154"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2103.03404","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2103.03404","pdf_url":"https://arxiv.org/pdf/2103.03404","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2103.03404","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5041835178","display_name":"Yihe Dong","orcid":"https://orcid.org/0000-0002-2390-485X","observed_orcids":["https://orcid.org/0000-0002-2390-485X"]},"institutions":[{"id":"https://openalex.org/I1291425158","display_name":"Google (United States)","ror":"https://ror.org/00njsd438","country_code":"US","type":"company","lineage":["https://openalex.org/I1291425158","https://openalex.org/I4210128969"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Dong, Yihe","raw_affiliation_strings":["Google,,,,,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google,,,,,","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5080236521","display_name":"Jean-Baptiste Cordonnier","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I5124864","display_name":"\u00c9cole Polytechnique F\u00e9d\u00e9rale de Lausanne","ror":"https://ror.org/02s376052","country_code":"CH","type":"education","lineage":["https://openalex.org/I2799323385","https://openalex.org/I5124864"]}],"countries":["CH"],"is_corresponding":false,"raw_author_name":"Cordonnier, Jean-Baptiste","raw_affiliation_strings":["EPFL","\u00c9cole Polytechnique F\u00e9d\u00e9rale de Lausanne,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"EPFL","institution_ids":["https://openalex.org/I5124864"]},{"raw_affiliation_string":"\u00c9cole Polytechnique F\u00e9d\u00e9rale de Lausanne,","institution_ids":["https://openalex.org/I5124864"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5033465803","display_name":"Andreas Loukas","orcid":"https://orcid.org/0000-0003-4866-1599","observed_orcids":["https://orcid.org/0000-0003-4866-1599"]},"institutions":[{"id":"https://openalex.org/I5124864","display_name":"\u00c9cole Polytechnique F\u00e9d\u00e9rale de Lausanne","ror":"https://ror.org/02s376052","country_code":"CH","type":"education","lineage":["https://openalex.org/I2799323385","https://openalex.org/I5124864"]}],"countries":["CH"],"is_corresponding":false,"raw_author_name":"Loukas, Andreas","raw_affiliation_strings":["EPFL","\u00c9cole Polytechnique F\u00e9d\u00e9rale de Lausanne,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"EPFL","institution_ids":["https://openalex.org/I5124864"]},{"raw_affiliation_string":"\u00c9cole Polytechnique F\u00e9d\u00e9rale de Lausanne,","institution_ids":["https://openalex.org/I5124864"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":243,"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/T10028","display_name":"Topic Modeling","score":0.5234000086784363,"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":0.5234000086784363,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.07999999821186066,"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/T10320","display_name":"Neural Networks and Applications","score":0.05079999938607216,"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/self-attention","display_name":"self-attention","score":0.9810000061988831},{"id":"https://openalex.org/keywords/attention-heads","display_name":"attention heads","score":0.7369999885559082},{"id":"https://openalex.org/keywords/inductive-bias","display_name":"inductive bias","score":0.7080000042915344},{"id":"https://openalex.org/keywords/transformer","display_name":"transformer models","score":0.6000000238418579},{"id":"https://openalex.org/keywords/skip-connections","display_name":"skip connections","score":0.8519999980926514},{"id":"https://openalex.org/keywords/multilayer-perceptron","display_name":"multilayer perceptron","score":0.8870000243186951}],"concepts":[{"id":"https://openalex.org/C60908668","wikidata":"https://www.wikidata.org/wiki/Q690207","display_name":"Perceptron","level":3,"score":0.6752723455429077},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6535718441009521},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.6012924909591675},{"id":"https://openalex.org/C75235859","wikidata":"https://www.wikidata.org/wiki/Q582659","display_name":"Exponential growth","level":2,"score":0.5464838147163391},{"id":"https://openalex.org/C164226766","wikidata":"https://www.wikidata.org/wiki/Q7293202","display_name":"Rank (graph theory)","level":2,"score":0.47045671939849854},{"id":"https://openalex.org/C2777303404","wikidata":"https://www.wikidata.org/wiki/Q759757","display_name":"Convergence (economics)","level":2,"score":0.4674433469772339},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.46623513102531433},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.44592371582984924},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3538253903388977},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.350835919380188},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.2793985605239868},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.27699965238571167},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.10561719536781311},{"id":"https://openalex.org/C165801399","wikidata":"https://www.wikidata.org/wiki/Q25428","display_name":"Voltage","level":2,"score":0.09667500853538513},{"id":"https://openalex.org/C31258907","wikidata":"https://www.wikidata.org/wiki/Q1301371","display_name":"Computer network","level":1,"score":0.08113709092140198},{"id":"https://openalex.org/C114614502","wikidata":"https://www.wikidata.org/wiki/Q76592","display_name":"Combinatorics","level":1,"score":0.08029067516326904},{"id":"https://openalex.org/C50522688","wikidata":"https://www.wikidata.org/wiki/Q189833","display_name":"Economic growth","level":1,"score":0.0},{"id":"https://openalex.org/C134306372","wikidata":"https://www.wikidata.org/wiki/Q7754","display_name":"Mathematical analysis","level":1,"score":0.0},{"id":"https://openalex.org/C86803240","wikidata":"https://www.wikidata.org/wiki/Q420","display_name":"Biology","level":0,"score":0.0},{"id":"https://openalex.org/C162324750","wikidata":"https://www.wikidata.org/wiki/Q8134","display_name":"Economics","level":0,"score":0.0},{"id":"https://openalex.org/C119599485","wikidata":"https://www.wikidata.org/wiki/Q43035","display_name":"Electrical engineering","level":1,"score":0.0},{"id":"https://openalex.org/C54355233","wikidata":"https://www.wikidata.org/wiki/Q7162","display_name":"Genetics","level":1,"score":0.0}],"mesh":[],"locations_count":4,"locations":[{"id":"pmh:oai:arXiv.org:2103.03404","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2103.03404","pdf_url":"https://arxiv.org/pdf/2103.03404","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"mag:3135593154","doi":null,"is_oa":true,"landing_page_url":"http://export.arxiv.org/pdf/2103.03404","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"arXiv (Cornell University)","raw_type":null},{"id":"pmh:oai:infoscience.epfl.ch:288790","doi":null,"is_oa":true,"landing_page_url":"http://infoscience.epfl.ch/record/288790","pdf_url":null,"source":{"id":"https://openalex.org/S4306400487","display_name":"Infoscience (Ecole Polytechnique F\u00e9d\u00e9rale de Lausanne)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I5124864","host_organization_name":"\u00c9cole Polytechnique F\u00e9d\u00e9rale de Lausanne","host_organization_lineage":["https://openalex.org/I5124864"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by-nc-nd","license_id":"https://openalex.org/licenses/cc-by-nc-nd","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"WoS","raw_type":"conference proceedings"},{"id":"doi:10.48550/arxiv.2103.03404","doi":"https://doi.org/10.48550/arxiv.2103.03404","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2103.03404","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2103.03404","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2103.03404","pdf_url":"https://arxiv.org/pdf/2103.03404","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[{"id":"https://openalex.org/G4430097687","display_name":null,"funder_award_id":"PZ00P2","funder_id":"https://openalex.org/F4320320924","funder_display_name":"Schweizerischer Nationalfonds zur F\u00f6rderung der Wissenschaftlichen Forschung"},{"id":"https://openalex.org/G4879078823","display_name":"Deep Learning for Graph-structured data","funder_award_id":"179981","funder_id":"https://openalex.org/F4320320924","funder_display_name":"Schweizerischer Nationalfonds zur F\u00f6rderung der Wissenschaftlichen Forschung"}],"funders":[{"id":"https://openalex.org/F4320306076","display_name":"National Science Foundation","ror":"https://ror.org/021nxhr62"},{"id":"https://openalex.org/F4320316545","display_name":"San Diego Supercomputer Center","ror":"https://ror.org/04mg3nk07"},{"id":"https://openalex.org/F4320320924","display_name":"Schweizerischer Nationalfonds zur F\u00f6rderung der Wissenschaftlichen Forschung","ror":"https://ror.org/00yjd3n13"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W3135593154.pdf","grobid_xml":"https://content.openalex.org/works/W3135593154.grobid-xml"},"referenced_works_count":35,"referenced_works":["https://openalex.org/W2089741025","https://openalex.org/W2194775991","https://openalex.org/W2305610840","https://openalex.org/W2525127255","https://openalex.org/W2591954064","https://openalex.org/W2626778328","https://openalex.org/W2866343820","https://openalex.org/W2885185669","https://openalex.org/W2892358082","https://openalex.org/W2908802752","https://openalex.org/W2937843571","https://openalex.org/W2950813464","https://openalex.org/W2962847566","https://openalex.org/W2963310665","https://openalex.org/W2963341956","https://openalex.org/W2963410064","https://openalex.org/W2964308564","https://openalex.org/W2970389371","https://openalex.org/W2975059944","https://openalex.org/W2978017171","https://openalex.org/W2994759459","https://openalex.org/W3021293129","https://openalex.org/W3030163527","https://openalex.org/W3033529678","https://openalex.org/W3033943443","https://openalex.org/W3034457371","https://openalex.org/W3034573343","https://openalex.org/W3036369012","https://openalex.org/W3038287124","https://openalex.org/W3045733172","https://openalex.org/W3085177480","https://openalex.org/W3104613728","https://openalex.org/W3112479704","https://openalex.org/W3119786062","https://openalex.org/W3169537799"],"related_works":["https://openalex.org/W7153067875","https://openalex.org/W4415798756","https://openalex.org/W7156752337","https://openalex.org/W3090952993","https://openalex.org/W6929270298","https://openalex.org/W4403364864","https://openalex.org/W7143046493","https://openalex.org/W3036369012","https://openalex.org/W4391590736","https://openalex.org/W7133245089"],"abstract_inverted_index":{"Attention-based":[0],"architectures":[1],"have":[2],"become":[3],"ubiquitous":[4],"in":[5],"machine":[6],"learning,":[7],"yet":[8],"our":[9],"understanding":[10],"of":[11,40,47,50,110],"the":[12,45,78,88,96,103],"reasons":[13],"for":[14],"their":[15,32],"effectiveness":[16],"remains":[17],"limited.":[18],"This":[19],"work":[20],"proposes":[21],"a":[22,38,48,63,84],"new":[23],"way":[24],"to":[25,83],"understand":[26],"self-attention":[27,61],"networks:":[28],"we":[29,58],"show":[30],"that":[31,60],"output":[33,79,97],"can":[34],"be":[35],"decomposed":[36],"into":[37],"sum":[39],"smaller":[41],"terms,":[42],"each":[43],"involving":[44],"operation":[46],"sequence":[49],"attention":[51],"heads":[52],"across":[53],"layers.":[54],"Using":[55],"this":[56],"decomposition,":[57],"prove":[59],"possesses":[62],"strong":[64],"inductive":[65],"bias":[66],"towards":[67],"\"token":[68],"uniformity\".":[69],"Specifically,":[70],"without":[71],"skip":[72,91],"connections":[73,92],"or":[74],"multi-layer":[75],"perceptrons":[76],"(MLPs),":[77],"converges":[80],"doubly":[81],"exponentially":[82],"rank-1":[85],"matrix.":[86],"On":[87],"other":[89],"hand,":[90],"and":[93],"MLPs":[94],"stop":[95],"from":[98],"degeneration.":[99],"Our":[100],"experiments":[101],"verify":[102],"identified":[104],"convergence":[105],"phenomena":[106],"on":[107],"different":[108],"variants":[109],"standard":[111],"transformer":[112],"architectures.":[113]},"counts_by_year":[{"year":2026,"cited_by_count":23},{"year":2025,"cited_by_count":41},{"year":2024,"cited_by_count":48},{"year":2023,"cited_by_count":57},{"year":2022,"cited_by_count":40},{"year":2021,"cited_by_count":32},{"year":2020,"cited_by_count":2}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2021-03-15T00:00:00"}