{"id":"https://openalex.org/W4224215744","doi":"https://doi.org/10.1371/journal.pone.0265621","title":"SensiMix: Sensitivity-Aware 8-bit index & 1-bit value mixed precision quantization for BERT compression","display_name":"SensiMix: Sensitivity-Aware 8-bit index & 1-bit value mixed precision quantization for BERT compression","publication_year":2022,"publication_date":"2022-04-18","ids":{"openalex":"https://openalex.org/W4224215744","doi":"https://doi.org/10.1371/journal.pone.0265621","pmid":"https://pubmed.ncbi.nlm.nih.gov/35436295","pmcid":"https://www.ncbi.nlm.nih.gov/pmc/articles/9015158"},"language":"en","primary_location":{"id":"doi:10.1371/journal.pone.0265621","doi":"https://doi.org/10.1371/journal.pone.0265621","is_oa":true,"landing_page_url":"https://doi.org/10.1371/journal.pone.0265621","pdf_url":"https://journals.plos.org/plosone/article/file?id=10.1371/journal.pone.0265621&type=printable","source":{"id":"https://openalex.org/S202381698","display_name":"PLoS ONE","issn_l":"1932-6203","issn":["1932-6203"],"is_oa":true,"is_in_doaj":true,"is_core":true,"listed_in":["anvur-class-a","anvur-scientific","cwts-core","doaj","dongbi-c","doyens","erih-plus","jufo-1","ki-jl-1","medline","norway-1","poland-100","russia-white-list-1","vabb-shw"],"host_organization":"https://openalex.org/P4310315706","host_organization_name":"Public Library of Science","host_organization_lineage":["https://openalex.org/P4310315706"],"host_organization_lineage_names":["Public Library of Science"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"PLOS ONE","raw_type":"journal-article"},"volume":null,"type":"article","indexed_in":["crossref","doaj","pubmed"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://journals.plos.org/plosone/article/file?id=10.1371/journal.pone.0265621&type=printable","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5039251882","display_name":"Tairen Piao","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I139264467","display_name":"Seoul National University","ror":"https://ror.org/04h9pn542","country_code":"KR","type":"education","lineage":["https://openalex.org/I139264467"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Tairen Piao","raw_affiliation_strings":["Seoul National University, Seoul, Republic of Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Seoul National University, Seoul, Republic of Korea","institution_ids":["https://openalex.org/I139264467"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5029401026","display_name":"Ikhyun Cho","orcid":"https://orcid.org/0009-0006-0360-256X","observed_orcids":["https://orcid.org/0009-0006-0360-256X"]},"institutions":[{"id":"https://openalex.org/I139264467","display_name":"Seoul National University","ror":"https://ror.org/04h9pn542","country_code":"KR","type":"education","lineage":["https://openalex.org/I139264467"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Ikhyun Cho","raw_affiliation_strings":["Seoul National University, Seoul, Republic of Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Seoul National University, Seoul, Republic of Korea","institution_ids":["https://openalex.org/I139264467"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5074561505","display_name":"U Kang","orcid":"https://orcid.org/0000-0002-8774-6950","observed_orcids":["https://orcid.org/0000-0002-8774-6950"]},"institutions":[{"id":"https://openalex.org/I139264467","display_name":"Seoul National University","ror":"https://ror.org/04h9pn542","country_code":"KR","type":"education","lineage":["https://openalex.org/I139264467"]}],"countries":["KR"],"is_corresponding":true,"raw_author_name":"U. Kang","raw_affiliation_strings":["Seoul National University, Seoul, Republic of Korea"],"raw_orcid":"https://orcid.org/0000-0002-8774-6950","affiliations":[{"raw_affiliation_string":"Seoul National University, Seoul, Republic of Korea","institution_ids":["https://openalex.org/I139264467"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":["https://openalex.org/A5074561505"],"corresponding_institution_ids":["https://openalex.org/I139264467"],"apc_list":{"value":1805,"currency":"USD","value_usd":1805},"apc_paid":{"value":1805,"currency":"USD","value_usd":1805},"fwci":0.7364,"has_fulltext":true,"cited_by_count":11,"citation_normalized_percentile":{"value":0.67069225,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":94,"max":97},"biblio":{"volume":"17","issue":"4","first_page":"e0265621","last_page":"e0265621"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.9940999746322632,"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":0.9940999746322632,"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/T10028","display_name":"Topic Modeling","score":0.001500000013038516,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.0007999999797903001,"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/mixed-precision-quantization","display_name":"mixed-precision quantization","score":0.9789999723434448},{"id":"https://openalex.org/keywords/model-compression","display_name":"model compression","score":0.46799999475479126},{"id":"https://openalex.org/keywords/inference-acceleration","display_name":"inference acceleration","score":0.6759999990463257},{"id":"https://openalex.org/keywords/glue-benchmark","display_name":"GLUE benchmark","score":0.44600000977516174}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7000885009765625},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.6834877729415894},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.5690866708755493},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.43069446086883545},{"id":"https://openalex.org/C207467116","wikidata":"https://www.wikidata.org/wiki/Q4385666","display_name":"Inverse","level":2,"score":0.42596757411956787},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.4239398241043091},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.22042453289031982},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.21191412210464478},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.1767238974571228},{"id":"https://openalex.org/C2524010","wikidata":"https://www.wikidata.org/wiki/Q8087","display_name":"Geometry","level":1,"score":0.0}],"mesh":[{"descriptor_ui":"D007802","descriptor_name":"Language","qualifier_ui":null,"qualifier_name":null,"is_major_topic":false},{"descriptor_ui":"D009323","descriptor_name":"Natural Language Processing","qualifier_ui":null,"qualifier_name":null,"is_major_topic":true},{"descriptor_ui":"D044962","descriptor_name":"Data Compression","qualifier_ui":null,"qualifier_name":null,"is_major_topic":true}],"locations_count":6,"locations":[{"id":"doi:10.1371/journal.pone.0265621","doi":"https://doi.org/10.1371/journal.pone.0265621","is_oa":true,"landing_page_url":"https://doi.org/10.1371/journal.pone.0265621","pdf_url":"https://journals.plos.org/plosone/article/file?id=10.1371/journal.pone.0265621&type=printable","source":{"id":"https://openalex.org/S202381698","display_name":"PLoS ONE","issn_l":"1932-6203","issn":["1932-6203"],"is_oa":true,"is_in_doaj":true,"is_core":true,"listed_in":["anvur-class-a","anvur-scientific","cwts-core","doaj","dongbi-c","doyens","erih-plus","jufo-1","ki-jl-1","medline","norway-1","poland-100","russia-white-list-1","vabb-shw"],"host_organization":"https://openalex.org/P4310315706","host_organization_name":"Public Library of Science","host_organization_lineage":["https://openalex.org/P4310315706"],"host_organization_lineage_names":["Public Library of Science"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"PLOS ONE","raw_type":"journal-article"},{"id":"pmh:oai:pubmedcentral.nih.gov:9015158","doi":null,"is_oa":true,"landing_page_url":"https://www.ncbi.nlm.nih.gov/pmc/articles/9015158","pdf_url":null,"source":{"id":"https://openalex.org/S2764455111","display_name":"PubMed Central","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I1299303238","host_organization_name":"National Institutes of Health","host_organization_lineage":["https://openalex.org/I1299303238"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"PLoS One","raw_type":"Text"},{"id":"pmid:35436295","doi":null,"is_oa":false,"landing_page_url":"https://pubmed.ncbi.nlm.nih.gov/35436295","pdf_url":null,"source":{"id":"https://openalex.org/S4306525036","display_name":"PubMed","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I1299303238","host_organization_name":"National Institutes of Health","host_organization_lineage":["https://openalex.org/I1299303238"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"PloS one","raw_type":"Journal Article"},{"id":"pmh:oai:doaj.org/article:304ad5f2f9db402baab25ae15d5e7b31","doi":null,"is_oa":true,"landing_page_url":"https://doaj.org/article/304ad5f2f9db402baab25ae15d5e7b31","pdf_url":null,"source":{"id":"https://openalex.org/S4306401280","display_name":"DOAJ (DOAJ: Directory of Open Access Journals)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by-sa","license_id":"https://openalex.org/licenses/cc-by-sa","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"PLoS ONE, Vol 17, Iss 4 (2022)","raw_type":"article"},{"id":"pmh:oai:doaj.org/article:b7d69c74a5ec446eb7a611caa8e6eb1c","doi":null,"is_oa":true,"landing_page_url":"https://doaj.org/article/b7d69c74a5ec446eb7a611caa8e6eb1c","pdf_url":null,"source":{"id":"https://openalex.org/S4306401280","display_name":"DOAJ (DOAJ: Directory of Open Access Journals)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by-sa","license_id":"https://openalex.org/licenses/cc-by-sa","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"PLoS ONE, Vol 17, Iss 4, p e0265621 (2022)","raw_type":"article"},{"id":"pmh:oai:s-space.snu.ac.kr:10371/183704","doi":null,"is_oa":true,"landing_page_url":"https://hdl.handle.net/10371/183704","pdf_url":null,"source":{"id":"https://openalex.org/S4306401345","display_name":"Seoul National University Open Repository (Seoul National University)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I139264467","host_organization_name":"Seoul National University","host_organization_lineage":["https://openalex.org/I139264467"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"","raw_type":"Article"}],"best_oa_location":{"id":"doi:10.1371/journal.pone.0265621","doi":"https://doi.org/10.1371/journal.pone.0265621","is_oa":true,"landing_page_url":"https://doi.org/10.1371/journal.pone.0265621","pdf_url":"https://journals.plos.org/plosone/article/file?id=10.1371/journal.pone.0265621&type=printable","source":{"id":"https://openalex.org/S202381698","display_name":"PLoS ONE","issn_l":"1932-6203","issn":["1932-6203"],"is_oa":true,"is_in_doaj":true,"is_core":true,"listed_in":["anvur-class-a","anvur-scientific","cwts-core","doaj","dongbi-c","doyens","erih-plus","jufo-1","ki-jl-1","medline","norway-1","poland-100","russia-white-list-1","vabb-shw"],"host_organization":"https://openalex.org/P4310315706","host_organization_name":"Public Library of Science","host_organization_lineage":["https://openalex.org/P4310315706"],"host_organization_lineage_names":["Public Library of Science"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"PLOS ONE","raw_type":"journal-article"},"sustainable_development_goals":[],"study_designs":[],"awards":[{"id":"https://openalex.org/G1152552616","display_name":null,"funder_award_id":"2021\u20130-01343","funder_id":"https://openalex.org/F4320335489","funder_display_name":"Institute for Information and Communications Technology Promotion"},{"id":"https://openalex.org/G1478288009","display_name":null,"funder_award_id":"2021-0-02068","funder_id":"https://openalex.org/F4320328359","funder_display_name":"Ministry of Science and ICT, South Korea"},{"id":"https://openalex.org/G2206812781","display_name":null,"funder_award_id":"2021-0-02068","funder_id":"https://openalex.org/F4320321292","funder_display_name":"Seoul National University"},{"id":"https://openalex.org/G3146912892","display_name":null,"funder_award_id":"2017-0-01772","funder_id":"https://openalex.org/F4320335489","funder_display_name":"Institute for Information and Communications Technology Promotion"},{"id":"https://openalex.org/G3705877861","display_name":null,"funder_award_id":"No.2021-0-02068","funder_id":"https://openalex.org/F4320328359","funder_display_name":"Ministry of Science and ICT, South Korea"},{"id":"https://openalex.org/G4643994530","display_name":null,"funder_award_id":"2021-0-02068","funder_id":"https://openalex.org/F4320335489","funder_display_name":"Institute for Information and Communications Technology Promotion"},{"id":"https://openalex.org/G5065061179","display_name":null,"funder_award_id":"2021-0-01343","funder_id":"https://openalex.org/F4320328359","funder_display_name":"Ministry of Science and ICT, South Korea"},{"id":"https://openalex.org/G5635159257","display_name":null,"funder_award_id":"2021-0-01343","funder_id":"https://openalex.org/F4320321292","funder_display_name":"Seoul National University"},{"id":"https://openalex.org/G5930330670","display_name":null,"funder_award_id":"No.2021-0-02068","funder_id":"https://openalex.org/F4320335489","funder_display_name":"Institute for Information and Communications Technology Promotion"},{"id":"https://openalex.org/G6593600532","display_name":null,"funder_award_id":"2020-0-00894","funder_id":"https://openalex.org/F4320335489","funder_display_name":"Institute for Information and Communications Technology Promotion"}],"funders":[{"id":"https://openalex.org/F4320321292","display_name":"Seoul National University","ror":"https://ror.org/04h9pn542"},{"id":"https://openalex.org/F4320328359","display_name":"Ministry of Science and ICT, South Korea","ror":"https://ror.org/01wpjm123"},{"id":"https://openalex.org/F4320335489","display_name":"Institute for Information and Communications Technology Promotion","ror":"https://ror.org/01g0hqq23"},{"id":"https://openalex.org/F4320336769","display_name":"Institute of Engineering Research, Seoul National University","ror":"https://ror.org/04scsat82"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4224215744.pdf","grobid_xml":"https://content.openalex.org/works/W4224215744.grobid-xml"},"referenced_works_count":27,"referenced_works":["https://openalex.org/W1821462560","https://openalex.org/W2285660444","https://openalex.org/W2560017826","https://openalex.org/W2626778328","https://openalex.org/W2766839578","https://openalex.org/W2777406049","https://openalex.org/W2886851211","https://openalex.org/W2923014074","https://openalex.org/W2937297214","https://openalex.org/W2948947170","https://openalex.org/W2953337107","https://openalex.org/W2965373594","https://openalex.org/W2969624041","https://openalex.org/W2973154008","https://openalex.org/W2978017171","https://openalex.org/W2979314664","https://openalex.org/W2997200074","https://openalex.org/W2997710335","https://openalex.org/W3089769400","https://openalex.org/W3089812288","https://openalex.org/W3094349784","https://openalex.org/W3113217532","https://openalex.org/W3184454880","https://openalex.org/W3191084366","https://openalex.org/W3202782831","https://openalex.org/W4233996382","https://openalex.org/W6768807518"],"related_works":["https://openalex.org/W4391306611","https://openalex.org/W3189437251","https://openalex.org/W4415957655","https://openalex.org/W3114304470","https://openalex.org/W4221154961","https://openalex.org/W2979314664","https://openalex.org/W3122752795","https://openalex.org/W7161763427","https://openalex.org/W7134110931","https://openalex.org/W2998183051"],"abstract_inverted_index":{"Given":[0],"a":[1,10,62,177],"pre-trained":[2],"BERT,":[3,24,94],"how":[4],"can":[5],"we":[6,55,131],"compress":[7],"it":[8],"to":[9,87,112,165],"fast":[11,129],"and":[12,48,83,90,123,138,143,181],"lightweight":[13,170],"one":[14],"while":[15,99],"maintaining":[16],"its":[17],"accuracy?":[18],"Pre-training":[19],"language":[20,33],"model,":[21,149],"such":[22],"as":[23],"is":[25],"effective":[26,168],"for":[27,128,141],"improving":[28],"the":[29,70,88,96,101,114,148,161,173,183],"performance":[30],"of":[31,44,72,75,93,147,179],"natural":[32],"processing":[34],"(NLP)":[35],"tasks.":[36],"However,":[37],"heavy":[38],"models":[39],"like":[40],"BERT":[41,65,163],"have":[42],"problems":[43],"large":[45],"memory":[46],"cost":[47],"long":[49],"inference":[50,184],"time.":[51],"In":[52],"this":[53],"paper,":[54],"propose":[56,106],"SensiMix":[57,77,159],"(Sensitivity-Aware":[58],"Mixed":[59],"Precision":[60],"Quantization),":[61],"novel":[63,108],"quantization-based":[64],"compression":[66,97],"method":[67],"that":[68,158],"considers":[69],"sensitivity":[71],"different":[73],"modules":[74],"BERT.":[76],"effectively":[78],"applies":[79],"8-bit":[80,142],"index":[81],"quantization":[82,86,145],"1-bit":[84,109,144],"value":[85],"sensitive":[89],"insensitive":[91],"parts":[92,146],"maximizing":[95],"rate":[98],"minimizing":[100],"accuracy":[102,115,191],"drop.":[103,192],"We":[104],"also":[105],"three":[107],"training":[110],"methods":[111],"minimize":[113],"drop:":[116],"Absolute":[117],"Binary":[118],"Weight":[119],"Regularization,":[120],"Prioritized":[121],"Training,":[122],"Inverse":[124],"Layer-wise":[125],"Fine-tuning.":[126],"Moreover,":[127],"inference,":[130],"apply":[132],"FP16":[133],"general":[134],"matrix":[135],"multiplication":[136],"(GEMM)":[137],"XNOR-Count":[139],"GEMM":[140],"respectively.":[150],"Experiments":[151],"on":[152],"four":[153],"GLUE":[154],"downstream":[155],"tasks":[156],"show":[157],"compresses":[160],"original":[162],"model":[164,174],"an":[166],"equally":[167],"but":[169],"one,":[171],"reducing":[172],"size":[175],"by":[176,186],"factor":[178],"8\u00d7":[180],"shrinking":[182],"time":[185],"around":[187],"80%":[188],"without":[189],"noticeable":[190]},"counts_by_year":[{"year":2026,"cited_by_count":2},{"year":2025,"cited_by_count":3},{"year":2024,"cited_by_count":2},{"year":2023,"cited_by_count":4}],"updated_date":"2026-10-31T12:09:11.096373","created_date":"2022-04-26T00:00:00"}