{"id":"https://openalex.org/W7108668758","doi":"https://doi.org/10.1109/tmm.2025.3639997","title":"EEformer: Early Exiting for Transformer With Global-Local Exits and Progressive Fine-Tuning","display_name":"EEformer: Early Exiting for Transformer With Global-Local Exits and Progressive Fine-Tuning","publication_year":2025,"publication_date":"2025-12-05","ids":{"openalex":"https://openalex.org/W7108668758","doi":"https://doi.org/10.1109/tmm.2025.3639997"},"language":null,"primary_location":{"id":"doi:10.1109/tmm.2025.3639997","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tmm.2025.3639997","pdf_url":null,"source":{"id":"https://openalex.org/S137030581","display_name":"IEEE Transactions on Multimedia","issn_l":"1520-9210","issn":["1520-9210","1941-0077"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319808","host_organization_name":"Institute of Electrical and Electronics Engineers","host_organization_lineage":["https://openalex.org/P4310319808"],"host_organization_lineage_names":["Institute of Electrical and Electronics Engineers"],"type":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"IEEE Transactions on Multimedia","raw_type":"journal-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":null,"display_name":"Guanyu Xu","orcid":"https://orcid.org/0000-0002-2771-9272"},"institutions":[{"id":"https://openalex.org/I125839683","display_name":"Beijing Institute of Technology","ror":"https://ror.org/01skt4w74","country_code":"CN","type":"education","lineage":["https://openalex.org/I125839683","https://openalex.org/I890469752"]}],"countries":["CN"],"is_corresponding":true,"raw_author_name":"Guanyu Xu","raw_affiliation_strings":["School of Information and Electronics, Beijing Institute of Technology, Beijing, China"],"raw_orcid":"https://orcid.org/0000-0002-2771-9272","affiliations":[{"raw_affiliation_string":"School of Information and Electronics, Beijing Institute of Technology, Beijing, China","institution_ids":["https://openalex.org/I125839683"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Jiawei Hao","orcid":"https://orcid.org/0009-0006-5050-4460"},"institutions":[{"id":"https://openalex.org/I125839683","display_name":"Beijing Institute of Technology","ror":"https://ror.org/01skt4w74","country_code":"CN","type":"education","lineage":["https://openalex.org/I125839683","https://openalex.org/I890469752"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Jiawei Hao","raw_affiliation_strings":["School of Information and Electronics, Beijing Institute of Technology, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0006-5050-4460","affiliations":[{"raw_affiliation_string":"School of Information and Electronics, Beijing Institute of Technology, Beijing, China","institution_ids":["https://openalex.org/I125839683"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Yong Luo","orcid":"https://orcid.org/0000-0002-2296-6370"},"institutions":[{"id":"https://openalex.org/I37461747","display_name":"Wuhan University","ror":"https://ror.org/033vjfk17","country_code":"CN","type":"education","lineage":["https://openalex.org/I37461747"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yong Luo","raw_affiliation_strings":["School of Computer Science, Wuhan University, Wuhan, China"],"raw_orcid":"https://orcid.org/0000-0002-2296-6370","affiliations":[{"raw_affiliation_string":"School of Computer Science, Wuhan University, Wuhan, China","institution_ids":["https://openalex.org/I37461747"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Li Shen","orcid":"https://orcid.org/0000-0001-5659-3464"},"institutions":[{"id":"https://openalex.org/I157773358","display_name":"Sun Yat-sen University","ror":"https://ror.org/0064kty71","country_code":"CN","type":"education","lineage":["https://openalex.org/I157773358"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Li Shen","raw_affiliation_strings":["School of Cyber Science and Technology, Shenzhen Campus of Sun Yat-sen University, Shenzhen, China"],"raw_orcid":"https://orcid.org/0000-0001-5659-3464","affiliations":[{"raw_affiliation_string":"School of Cyber Science and Technology, Shenzhen Campus of Sun Yat-sen University, Shenzhen, China","institution_ids":["https://openalex.org/I157773358"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Han Hu","orcid":"https://orcid.org/0000-0001-7532-0496"},"institutions":[{"id":"https://openalex.org/I125839683","display_name":"Beijing Institute of Technology","ror":"https://ror.org/01skt4w74","country_code":"CN","type":"education","lineage":["https://openalex.org/I125839683","https://openalex.org/I890469752"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Han Hu","raw_affiliation_strings":["School of Information and Electronics, Beijing Institute of Technology, Beijing, China"],"raw_orcid":"https://orcid.org/0000-0001-7532-0496","affiliations":[{"raw_affiliation_string":"School of Information and Electronics, Beijing Institute of Technology, Beijing, China","institution_ids":["https://openalex.org/I125839683"]}]},{"author_position":"last","author":{"id":null,"display_name":"Dan Zeng","orcid":"https://orcid.org/0000-0003-1300-1769"},"institutions":[{"id":"https://openalex.org/I113940042","display_name":"Shanghai University","ror":"https://ror.org/006teas31","country_code":"CN","type":"education","lineage":["https://openalex.org/I113940042"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Dan Zeng","raw_affiliation_strings":["School of Communication and Information Engineering, Shanghai University, Shanghai, China"],"raw_orcid":"https://orcid.org/0000-0003-1300-1769","affiliations":[{"raw_affiliation_string":"School of Communication and Information Engineering, Shanghai University, Shanghai, China","institution_ids":["https://openalex.org/I113940042"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":6,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I125839683"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.57705183,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":"28","issue":null,"first_page":"1352","last_page":"1364"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.19429999589920044,"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":0.19429999589920044,"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/T10273","display_name":"IoT and Edge/Fog Computing","score":0.06700000166893005,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/T10714","display_name":"Software-Defined Networks and 5G","score":0.03590000048279762,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/transformer","display_name":"Transformer","score":0.6657999753952026},{"id":"https://openalex.org/keywords/software-deployment","display_name":"Software deployment","score":0.5684999823570251},{"id":"https://openalex.org/keywords/edge-device","display_name":"Edge device","score":0.5514000058174133},{"id":"https://openalex.org/keywords/acceleration","display_name":"Acceleration","score":0.49410000443458557},{"id":"https://openalex.org/keywords/artificial-neural-network","display_name":"Artificial neural network","score":0.48660001158714294},{"id":"https://openalex.org/keywords/construct","display_name":"Construct (python library)","score":0.46650001406669617},{"id":"https://openalex.org/keywords/enhanced-data-rates-for-gsm-evolution","display_name":"Enhanced Data Rates for GSM Evolution","score":0.40310001373291016},{"id":"https://openalex.org/keywords/degradation","display_name":"Degradation (telecommunications)","score":0.33970001339912415},{"id":"https://openalex.org/keywords/edge-computing","display_name":"Edge computing","score":0.3393000066280365}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8756999969482422},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.6657999753952026},{"id":"https://openalex.org/C105339364","wikidata":"https://www.wikidata.org/wiki/Q2297740","display_name":"Software deployment","level":2,"score":0.5684999823570251},{"id":"https://openalex.org/C138236772","wikidata":"https://www.wikidata.org/wiki/Q25098575","display_name":"Edge device","level":3,"score":0.5514000058174133},{"id":"https://openalex.org/C117896860","wikidata":"https://www.wikidata.org/wiki/Q11376","display_name":"Acceleration","level":2,"score":0.49410000443458557},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.48660001158714294},{"id":"https://openalex.org/C2780801425","wikidata":"https://www.wikidata.org/wiki/Q5164392","display_name":"Construct (python library)","level":2,"score":0.46650001406669617},{"id":"https://openalex.org/C162307627","wikidata":"https://www.wikidata.org/wiki/Q204833","display_name":"Enhanced Data Rates for GSM Evolution","level":2,"score":0.40310001373291016},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.39239999651908875},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.38019999861717224},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.36320000886917114},{"id":"https://openalex.org/C2779679103","wikidata":"https://www.wikidata.org/wiki/Q5251805","display_name":"Degradation (telecommunications)","level":2,"score":0.33970001339912415},{"id":"https://openalex.org/C2778456923","wikidata":"https://www.wikidata.org/wiki/Q5337692","display_name":"Edge computing","level":3,"score":0.3393000066280365},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.33469998836517334},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.31450000405311584},{"id":"https://openalex.org/C9417928","wikidata":"https://www.wikidata.org/wiki/Q1070689","display_name":"Image processing","level":3,"score":0.3068999946117401},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.3061000108718872},{"id":"https://openalex.org/C52622490","wikidata":"https://www.wikidata.org/wiki/Q1026626","display_name":"Feature extraction","level":2,"score":0.3028999865055084},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.2978000044822693},{"id":"https://openalex.org/C104267543","wikidata":"https://www.wikidata.org/wiki/Q208163","display_name":"Signal processing","level":3,"score":0.2874999940395355},{"id":"https://openalex.org/C2778915421","wikidata":"https://www.wikidata.org/wiki/Q3643177","display_name":"Performance improvement","level":2,"score":0.2824999988079071},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.27300000190734863},{"id":"https://openalex.org/C147168706","wikidata":"https://www.wikidata.org/wiki/Q1457734","display_name":"Recurrent neural network","level":3,"score":0.26820001006126404},{"id":"https://openalex.org/C23725684","wikidata":"https://www.wikidata.org/wiki/Q616377","display_name":"Maintenance engineering","level":2,"score":0.2662999927997589},{"id":"https://openalex.org/C2984842247","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep neural networks","level":3,"score":0.2630000114440918},{"id":"https://openalex.org/C44154836","wikidata":"https://www.wikidata.org/wiki/Q45045","display_name":"Simulation","level":1,"score":0.2614000141620636},{"id":"https://openalex.org/C147764199","wikidata":"https://www.wikidata.org/wiki/Q6865248","display_name":"Minification","level":2,"score":0.25040000677108765}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/tmm.2025.3639997","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tmm.2025.3639997","pdf_url":null,"source":{"id":"https://openalex.org/S137030581","display_name":"IEEE Transactions on Multimedia","issn_l":"1520-9210","issn":["1520-9210","1941-0077"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319808","host_organization_name":"Institute of Electrical and Electronics Engineers","host_organization_lineage":["https://openalex.org/P4310319808"],"host_organization_lineage_names":["Institute of Electrical and Electronics Engineers"],"type":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"IEEE Transactions on Multimedia","raw_type":"journal-article"}],"best_oa_location":null,"sustainable_development_goals":[{"score":0.4100439250469208,"id":"https://metadata.un.org/sdg/8","display_name":"Decent work and economic growth"}],"awards":[{"id":"https://openalex.org/G1058125861","display_name":null,"funder_award_id":"62372284","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"},{"id":"https://openalex.org/G5019704030","display_name":null,"funder_award_id":"U2336211","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"},{"id":"https://openalex.org/G5548306916","display_name":null,"funder_award_id":"92467206","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"},{"id":"https://openalex.org/G6704236761","display_name":null,"funder_award_id":"62202283","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"}],"funders":[{"id":"https://openalex.org/F4320321001","display_name":"National Natural Science Foundation of China","ror":"https://ror.org/01h0zpd94"}],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":36,"referenced_works":["https://openalex.org/W12634471","https://openalex.org/W1582774210","https://openalex.org/W2108598243","https://openalex.org/W2251939518","https://openalex.org/W2593833795","https://openalex.org/W2896457183","https://openalex.org/W2923014074","https://openalex.org/W2962677625","https://openalex.org/W2963748441","https://openalex.org/W2963846996","https://openalex.org/W2979826702","https://openalex.org/W2981812042","https://openalex.org/W2982605490","https://openalex.org/W3035030897","https://openalex.org/W3035038672","https://openalex.org/W3038012435","https://openalex.org/W3094502228","https://openalex.org/W3137609883","https://openalex.org/W3138516171","https://openalex.org/W3154971029","https://openalex.org/W3176017841","https://openalex.org/W3211113236","https://openalex.org/W4281571140","https://openalex.org/W4285059684","https://openalex.org/W4285601701","https://openalex.org/W4287887900","https://openalex.org/W4313318649","https://openalex.org/W4382239116","https://openalex.org/W4385570210","https://openalex.org/W4385571084","https://openalex.org/W4385768057","https://openalex.org/W4385958936","https://openalex.org/W4386065441","https://openalex.org/W4393147201","https://openalex.org/W4405231929","https://openalex.org/W4411245047"],"related_works":[],"abstract_inverted_index":{"Recently,":[0],"the":[1,42,65,86,96,109,112,120,140,143,148,159,170],"efficient":[2],"deployment":[3],"and":[4,36,78,134,162],"acceleration":[5],"of":[6,111,142],"transformer-based":[7],"pre-trained":[8],"models":[9],"(TPMs)":[10],"on":[11,131,158,169],"resource-constrained":[12],"edge":[13],"devices":[14],"for":[15,95],"multimedia":[16],"services":[17],"have":[18],"gained":[19],"significant":[20],"interest.":[21],"Although":[22],"early":[23,66],"exiting":[24,67],"is":[25],"a":[26,51,91,101],"feasible":[27],"solution,":[28],"it":[29],"may":[30],"lead":[31],"to":[32,41,63,106,119],"extra":[33],"computational":[34],"cost":[35],"substantial":[37],"performance":[38,117,157,168],"degradation":[39],"compared":[40],"original":[43,121],"models.":[44],"To":[45],"tackle":[46],"these":[47],"issues,":[48],"we":[49,99],"propose":[50,100],"framework":[52,150],"termed":[53],"EEformer,":[54],"which":[55],"incorporates":[56],"global-local":[57],"heads":[58],"(GLHs)":[59],"into":[60],"intermediate":[61],"layers":[62],"construct":[64],"dynamic":[68],"neural":[69],"network":[70],"(EDNN).":[71],"The":[72],"GLH":[73],"can":[74],"efficiently":[75],"extract":[76],"global":[77],"local":[79],"information":[80],"from":[81],"hidden":[82],"states":[83],"produced":[84],"by":[85],"backbone":[87],"layer,":[88],"thereby":[89],"achieving":[90],"better":[92],"performance-efficiency":[93],"trade-off":[94],"EDNN.":[97],"Moreover,":[98],"novel":[102],"progressive":[103],"fine-tuning":[104,125],"strategy":[105],"steadily":[107],"improve":[108],"efficiency":[110],"EDNN":[113],"while":[114,154,165],"maintaining":[115,155,166],"its":[116],"comparable":[118],"mode":[122],"through":[123],"three":[124],"stages.":[126],"We":[127],"conduct":[128],"extensive":[129],"experiments":[130],"image":[132],"classification":[133],"natural":[135],"language":[136],"processing":[137],"tasks,":[138],"demonstrating":[139],"superiority":[141],"proposed":[144,149],"framework.":[145],"In":[146],"particular,":[147],"achieves":[151],"1.87\u00d7":[152],"speed-up":[153,164],"99.0%":[156],"CIFAR-100":[160],"dataset,":[161],"3.05\u00d7":[163],"98.5%":[167],"SST-2":[171],"dataset.":[172]},"counts_by_year":[],"updated_date":"2026-03-09T07:00:12.390032","created_date":"2025-12-05T00:00:00"}