{"id":"https://openalex.org/W4206281850","doi":"https://doi.org/10.1162/tacl_a_00436","title":"Differentiable Subset Pruning of Transformer Heads","display_name":"Differentiable Subset Pruning of Transformer Heads","publication_year":2021,"publication_date":"2021-01-01","ids":{"openalex":"https://openalex.org/W4206281850","doi":"https://doi.org/10.1162/tacl_a_00436"},"language":"en","primary_location":{"id":"doi:10.1162/tacl_a_00436","is_oa":true,"landing_page_url":"https://doi.org/10.1162/tacl_a_00436","pdf_url":"https://direct.mit.edu/tacl/article-pdf/doi/10.1162/tacl_a_00436/1979279/tacl_a_00436.pdf","source":{"id":"https://openalex.org/S2729999759","display_name":"Transactions of the Association for Computational Linguistics","issn_l":"2307-387X","issn":["2307-387X"],"is_oa":true,"is_in_doaj":true,"is_core":true,"host_organization":"https://openalex.org/P4310320244","host_organization_name":"Association for Computational Linguistics","host_organization_lineage":["https://openalex.org/P4310320244"],"host_organization_lineage_names":["Association for Computational Linguistics"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Transactions of the Association for Computational Linguistics","raw_type":"journal-article"},"type":"article","indexed_in":["crossref","doaj"],"open_access":{"is_oa":true,"oa_status":"diamond","oa_url":"https://direct.mit.edu/tacl/article-pdf/doi/10.1162/tacl_a_00436/1979279/tacl_a_00436.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5017343435","display_name":"Jiaoda Li","orcid":null},"institutions":[{"id":"https://openalex.org/I35440088","display_name":"ETH Zurich","ror":"https://ror.org/05a28rw58","country_code":"CH","type":"education","lineage":["https://openalex.org/I2799323385","https://openalex.org/I35440088"]}],"countries":["CH"],"is_corresponding":true,"raw_author_name":"Jiaoda Li","raw_affiliation_strings":["ETH Z\u00fcrich, Switzerland. jiaoda.li@inf.ethz.ch"],"affiliations":[{"raw_affiliation_string":"ETH Z\u00fcrich, Switzerland. jiaoda.li@inf.ethz.ch","institution_ids":["https://openalex.org/I35440088"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5061951606","display_name":"Ryan Cotterell","orcid":"https://orcid.org/0000-0003-4080-1833"},"institutions":[{"id":"https://openalex.org/I241749","display_name":"University of Cambridge","ror":"https://ror.org/013meh722","country_code":"GB","type":"education","lineage":["https://openalex.org/I241749"]},{"id":"https://openalex.org/I35440088","display_name":"ETH Zurich","ror":"https://ror.org/05a28rw58","country_code":"CH","type":"education","lineage":["https://openalex.org/I2799323385","https://openalex.org/I35440088"]}],"countries":["CH","GB"],"is_corresponding":true,"raw_author_name":"Ryan Cotterell","raw_affiliation_strings":["ETH Z\u00fcrich, Switzerland","University of Cambridge, UK. ryan.cotterell@inf.ethz.ch"],"affiliations":[{"raw_affiliation_string":"ETH Z\u00fcrich, Switzerland","institution_ids":["https://openalex.org/I35440088"]},{"raw_affiliation_string":"University of Cambridge, UK. ryan.cotterell@inf.ethz.ch","institution_ids":["https://openalex.org/I241749"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5002316432","display_name":"Mrinmaya Sachan","orcid":"https://orcid.org/0000-0001-8787-8681"},"institutions":[{"id":"https://openalex.org/I35440088","display_name":"ETH Zurich","ror":"https://ror.org/05a28rw58","country_code":"CH","type":"education","lineage":["https://openalex.org/I2799323385","https://openalex.org/I35440088"]}],"countries":["CH"],"is_corresponding":true,"raw_author_name":"Mrinmaya Sachan","raw_affiliation_strings":["ETH Z\u00fcrich, Switzerland. mrinmaya.sachan@inf.ethz.ch"],"affiliations":[{"raw_affiliation_string":"ETH Z\u00fcrich, Switzerland. mrinmaya.sachan@inf.ethz.ch","institution_ids":["https://openalex.org/I35440088"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":3,"corresponding_author_ids":["https://openalex.org/A5002316432","https://openalex.org/A5017343435","https://openalex.org/A5061951606"],"corresponding_institution_ids":["https://openalex.org/I241749","https://openalex.org/I35440088"],"apc_list":null,"apc_paid":null,"fwci":2.7188,"has_fulltext":true,"cited_by_count":38,"citation_normalized_percentile":{"value":0.9204718,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":96,"max":100},"biblio":{"volume":"9","issue":null,"first_page":"1442","last_page":"1459"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.9944999814033508,"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":0.9944999814033508,"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":0.984000027179718,"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/T10601","display_name":"Handwritten Text Recognition Techniques","score":0.9754999876022339,"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":0.8350197672843933},{"id":"https://openalex.org/keywords/differentiable-function","display_name":"Differentiable function","score":0.7549800872802734},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.6686335206031799},{"id":"https://openalex.org/keywords/pruning","display_name":"Pruning","score":0.6251684427261353},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5913239121437073},{"id":"https://openalex.org/keywords/machine-translation","display_name":"Machine translation","score":0.48892632126808167},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.4775006175041199},{"id":"https://openalex.org/keywords/gradient-descent","display_name":"Gradient descent","score":0.43076226115226746},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.4111217260360718},{"id":"https://openalex.org/keywords/machine-learning","display_name":"Machine learning","score":0.3859228193759918},{"id":"https://openalex.org/keywords/artificial-neural-network","display_name":"Artificial neural network","score":0.11731579899787903},{"id":"https://openalex.org/keywords/mathematics","display_name":"Mathematics","score":0.11045783758163452}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8350197672843933},{"id":"https://openalex.org/C202615002","wikidata":"https://www.wikidata.org/wiki/Q783507","display_name":"Differentiable function","level":2,"score":0.7549800872802734},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.6686335206031799},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.6251684427261353},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5913239121437073},{"id":"https://openalex.org/C203005215","wikidata":"https://www.wikidata.org/wiki/Q79798","display_name":"Machine translation","level":2,"score":0.48892632126808167},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4775006175041199},{"id":"https://openalex.org/C153258448","wikidata":"https://www.wikidata.org/wiki/Q1199743","display_name":"Gradient descent","level":3,"score":0.43076226115226746},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.4111217260360718},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3859228193759918},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.11731579899787903},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.11045783758163452},{"id":"https://openalex.org/C86803240","wikidata":"https://www.wikidata.org/wiki/Q420","display_name":"Biology","level":0,"score":0.0},{"id":"https://openalex.org/C6557445","wikidata":"https://www.wikidata.org/wiki/Q173113","display_name":"Agronomy","level":1,"score":0.0},{"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},{"id":"https://openalex.org/C134306372","wikidata":"https://www.wikidata.org/wiki/Q7754","display_name":"Mathematical analysis","level":1,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.1162/tacl_a_00436","is_oa":true,"landing_page_url":"https://doi.org/10.1162/tacl_a_00436","pdf_url":"https://direct.mit.edu/tacl/article-pdf/doi/10.1162/tacl_a_00436/1979279/tacl_a_00436.pdf","source":{"id":"https://openalex.org/S2729999759","display_name":"Transactions of the Association for Computational Linguistics","issn_l":"2307-387X","issn":["2307-387X"],"is_oa":true,"is_in_doaj":true,"is_core":true,"host_organization":"https://openalex.org/P4310320244","host_organization_name":"Association for Computational Linguistics","host_organization_lineage":["https://openalex.org/P4310320244"],"host_organization_lineage_names":["Association for Computational Linguistics"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Transactions of the Association for Computational Linguistics","raw_type":"journal-article"},{"id":"pmh:oai:doaj.org/article:8d9f4a5039f549b3bd16d4d7d11a69de","is_oa":true,"landing_page_url":"https://doaj.org/article/8d9f4a5039f549b3bd16d4d7d11a69de","pdf_url":null,"source":{"id":"https://openalex.org/S112646816","display_name":"SHILAP Revista de lepidopterolog\u00eda","issn_l":"0300-5267","issn":["0300-5267","2340-4078"],"is_oa":true,"is_in_doaj":true,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"journal"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"Transactions of the Association for Computational Linguistics, Vol 9, Pp 1442-1459 (2021)","raw_type":"article"}],"best_oa_location":{"id":"doi:10.1162/tacl_a_00436","is_oa":true,"landing_page_url":"https://doi.org/10.1162/tacl_a_00436","pdf_url":"https://direct.mit.edu/tacl/article-pdf/doi/10.1162/tacl_a_00436/1979279/tacl_a_00436.pdf","source":{"id":"https://openalex.org/S2729999759","display_name":"Transactions of the Association for Computational Linguistics","issn_l":"2307-387X","issn":["2307-387X"],"is_oa":true,"is_in_doaj":true,"is_core":true,"host_organization":"https://openalex.org/P4310320244","host_organization_name":"Association for Computational Linguistics","host_organization_lineage":["https://openalex.org/P4310320244"],"host_organization_lineage_names":["Association for Computational Linguistics"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Transactions of the Association for Computational Linguistics","raw_type":"journal-article"},"sustainable_development_goals":[{"score":0.46000000834465027,"id":"https://metadata.un.org/sdg/4","display_name":"Quality Education"}],"awards":[{"id":"https://openalex.org/G1158548905","display_name":"Representation Learning for Arbitrarily Long Richly Formatted Multimedia Documents","funder_award_id":"201009","funder_id":"https://openalex.org/F4320320924","funder_display_name":"Schweizerischer Nationalfonds zur F\u00f6rderung der Wissenschaftlichen Forschung"},{"id":"https://openalex.org/G1973324782","display_name":"Agonie religieuse et transfiguration politique du jans\u00e9nisme au XVIIIe si\u00e8cle.","funder_award_id":"20100","funder_id":"https://openalex.org/F4320320924","funder_display_name":"Schweizerischer Nationalfonds zur F\u00f6rderung der Wissenschaftlichen Forschung"}],"funders":[{"id":"https://openalex.org/F4320320924","display_name":"Schweizerischer Nationalfonds zur F\u00f6rderung der Wissenschaftlichen Forschung","ror":"https://ror.org/00yjd3n13"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4206281850.pdf","grobid_xml":"https://content.openalex.org/works/W4206281850.grobid-xml"},"referenced_works_count":76,"referenced_works":["https://openalex.org/W1631063262","https://openalex.org/W1994530392","https://openalex.org/W2095705004","https://openalex.org/W2119885577","https://openalex.org/W2134445663","https://openalex.org/W2242818861","https://openalex.org/W2346626577","https://openalex.org/W2559655401","https://openalex.org/W2612690371","https://openalex.org/W2783873922","https://openalex.org/W2797569913","https://openalex.org/W2914584698","https://openalex.org/W2917458986","https://openalex.org/W2946794439","https://openalex.org/W2947644971","https://openalex.org/W2948947170","https://openalex.org/W2950656546","https://openalex.org/W2963021447","https://openalex.org/W2963069394","https://openalex.org/W2963363373","https://openalex.org/W2963382930","https://openalex.org/W2964233199","https://openalex.org/W2964554227","https://openalex.org/W2970295111","https://openalex.org/W2972324944","https://openalex.org/W2979826702","https://openalex.org/W2990138404","https://openalex.org/W3004346089","https://openalex.org/W3015982254","https://openalex.org/W3034230673","https://openalex.org/W3038012435","https://openalex.org/W3102446692","https://openalex.org/W3102519438","https://openalex.org/W3103754749","https://openalex.org/W3136363192","https://openalex.org/W4238634189","https://openalex.org/W4288631803","https://openalex.org/W4301409532","https://openalex.org/W6617145748","https://openalex.org/W6638836233","https://openalex.org/W6674330103","https://openalex.org/W6677103964","https://openalex.org/W6678251536","https://openalex.org/W6679434410","https://openalex.org/W6681015464","https://openalex.org/W6690026940","https://openalex.org/W6697501834","https://openalex.org/W6724998850","https://openalex.org/W6725543821","https://openalex.org/W6726275242","https://openalex.org/W6729448088","https://openalex.org/W6730091202","https://openalex.org/W6732814185","https://openalex.org/W6734621010","https://openalex.org/W6735799286","https://openalex.org/W6737236263","https://openalex.org/W6739901393","https://openalex.org/W6739917289","https://openalex.org/W6743087176","https://openalex.org/W6745148473","https://openalex.org/W6745499552","https://openalex.org/W6748171661","https://openalex.org/W6751979845","https://openalex.org/W6755207826","https://openalex.org/W6757883768","https://openalex.org/W6760167989","https://openalex.org/W6761164592","https://openalex.org/W6762945437","https://openalex.org/W6762947928","https://openalex.org/W6763701032","https://openalex.org/W6768080748","https://openalex.org/W6775432837","https://openalex.org/W6777017071","https://openalex.org/W6778883912","https://openalex.org/W6791096077","https://openalex.org/W6802318714"],"related_works":["https://openalex.org/W3140942413","https://openalex.org/W4303874710","https://openalex.org/W2222501782","https://openalex.org/W4316672576","https://openalex.org/W4281712593","https://openalex.org/W3176018525","https://openalex.org/W3033942572","https://openalex.org/W4281656837","https://openalex.org/W3153803265","https://openalex.org/W4292641180"],"abstract_inverted_index":{"Abstract":[0],"Multi-head":[1],"attention,":[2],"a":[3,31,38,72,94],"collection":[4],"of":[5,15,34,53,101,140],"several":[6],"attention":[7,41],"mechanisms":[8],"that":[9,30,61,77,125],"independently":[10],"attend":[11],"to":[12,59],"different":[13],"parts":[14],"the":[16,19,23,35,51,54,99,141],"input,":[17],"is":[18],"key":[20],"ingredient":[21],"in":[22,37,67],"Transformer.":[24],"Recent":[25],"work":[26,70],"has":[27],"shown,":[28],"however,":[29],"large":[32],"proportion":[33],"heads":[36],"Transformer\u2019s":[39],"multi-head":[40],"mechanism":[42],"can":[43],"be":[44],"safely":[45],"pruned":[46],"away":[47],"without":[48],"significantly":[49],"harming":[50],"performance":[52],"model;":[55],"such":[56],"pruning":[57,75,128],"leads":[58],"models":[60],"are":[62,107],"noticeably":[63],"smaller":[64],"and":[65,91,120],"faster":[66],"practice.":[68],"Our":[69],"introduces":[71],"new":[73],"head":[74,88],"technique":[76],"we":[78,123],"term":[79],"differentiable":[80,126],"subset":[81,127],"pruning.":[82],"ntuitively,":[83],"our":[84],"method":[85],"learns":[86],"per-":[87],"importance":[89,105],"variables":[90,106],"then":[92],"enforces":[93],"user-specified":[95],"hard":[96],"constraint":[97],"on":[98,116],"number":[100],"unpruned":[102],"heads.":[103],"he":[104],"learned":[108],"via":[109],"stochastic":[110],"gradient":[111],"descent.":[112],"e":[113],"conduct":[114],"experiments":[115],"natural":[117],"language":[118],"inference":[119],"machine":[121],"translation;":[122],"show":[124],"performs":[129],"comparably":[130],"or":[131],"better":[132],"than":[133],"previous":[134],"works":[135],"while":[136],"offering":[137],"precise":[138],"control":[139],"sparsity":[142],"level.1":[143]},"counts_by_year":[{"year":2026,"cited_by_count":1},{"year":2025,"cited_by_count":9},{"year":2024,"cited_by_count":6},{"year":2023,"cited_by_count":15},{"year":2022,"cited_by_count":7}],"updated_date":"2026-04-10T15:06:20.359241","created_date":"2025-10-10T00:00:00"}