{"id":"https://openalex.org/W4416016882","doi":"https://doi.org/10.1145/3746252.3761011","title":"Unplug and Play Language Models: Decomposing Experts in Language Models at Inference Time","display_name":"Unplug and Play Language Models: Decomposing Experts in Language Models at Inference Time","publication_year":2025,"publication_date":"2025-11-07","ids":{"openalex":"https://openalex.org/W4416016882","doi":"https://doi.org/10.1145/3746252.3761011"},"language":null,"primary_location":{"id":"doi:10.1145/3746252.3761011","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3746252.3761011","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 34th ACM International Conference on Information and Knowledge Management","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://doi.org/10.1145/3746252.3761011","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5036753528","display_name":"Nakyeong Yang","orcid":"https://orcid.org/0000-0002-2196-5149"},"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":"Nakyeong Yang","raw_affiliation_strings":["Seoul National University, Seoul, Republic of Korea"],"raw_orcid":"https://orcid.org/0000-0002-2196-5149","affiliations":[{"raw_affiliation_string":"Seoul National University, Seoul, Republic of Korea","institution_ids":["https://openalex.org/I139264467"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Jiwon Moon","orcid":"https://orcid.org/0009-0003-1863-4220"},"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":"Jiwon Moon","raw_affiliation_strings":["Seoul National University, Seoul, Republic of Korea"],"raw_orcid":"https://orcid.org/0009-0003-1863-4220","affiliations":[{"raw_affiliation_string":"Seoul National University, Seoul, Republic of Korea","institution_ids":["https://openalex.org/I139264467"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Junseok Kim","orcid":"https://orcid.org/0009-0004-1531-1415"},"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":"Junseok Kim","raw_affiliation_strings":["Seoul National University, Seoul, Republic of Korea"],"raw_orcid":"https://orcid.org/0009-0004-1531-1415","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/A5088353147","display_name":"Yunah Jang","orcid":"https://orcid.org/0000-0002-2805-7530"},"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":"Yunah Jang","raw_affiliation_strings":["Seoul National University, Seoul, Republic of Korea"],"raw_orcid":"https://orcid.org/0000-0002-2805-7530","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/A5077832834","display_name":"Kyomin Jung","orcid":"https://orcid.org/0000-0003-2547-7051"},"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":"Kyomin Jung","raw_affiliation_strings":["Seoul National University, Seoul, Republic of Korea"],"raw_orcid":"https://orcid.org/0000-0003-2547-7051","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":5,"corresponding_author_ids":["https://openalex.org/A5036753528"],"corresponding_institution_ids":["https://openalex.org/I139264467"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.16762227,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"3805","last_page":"3813"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.2533999979496002,"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":0.2533999979496002,"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/T11636","display_name":"Artificial Intelligence in Healthcare and Education","score":0.19660000503063202,"subfield":{"id":"https://openalex.org/subfields/2718","display_name":"Health Informatics"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}},{"id":"https://openalex.org/T11704","display_name":"Mobile Crowdsensing and Crowdsourcing","score":0.09600000083446503,"subfield":{"id":"https://openalex.org/subfields/1706","display_name":"Computer Science Applications"},"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/inference","display_name":"Inference","score":0.7565000057220459},{"id":"https://openalex.org/keywords/task","display_name":"Task (project management)","score":0.7441999912261963},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.6658999919891357},{"id":"https://openalex.org/keywords/set","display_name":"Set (abstract data type)","score":0.6173999905586243},{"id":"https://openalex.org/keywords/pruning","display_name":"Pruning","score":0.5479000210762024},{"id":"https://openalex.org/keywords/natural-language","display_name":"Natural language","score":0.4602999985218048},{"id":"https://openalex.org/keywords/overhead","display_name":"Overhead (engineering)","score":0.4537999927997589},{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.44110000133514404}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8259000182151794},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.7565000057220459},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.7441999912261963},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.6658999919891357},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.6388000249862671},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.6173999905586243},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.5479000210762024},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.5291000008583069},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.5252000093460083},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.4602999985218048},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.4537999927997589},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.44110000133514404},{"id":"https://openalex.org/C44291984","wikidata":"https://www.wikidata.org/wiki/Q1074173","display_name":"Question answering","level":2,"score":0.415800005197525},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.4131999909877777},{"id":"https://openalex.org/C124681953","wikidata":"https://www.wikidata.org/wiki/Q339062","display_name":"Decomposition","level":2,"score":0.3504999876022339},{"id":"https://openalex.org/C179603123","wikidata":"https://www.wikidata.org/wiki/Q1941921","display_name":"Modeling language","level":3,"score":0.32420000433921814},{"id":"https://openalex.org/C2779439875","wikidata":"https://www.wikidata.org/wiki/Q1078276","display_name":"Natural language understanding","level":3,"score":0.3009999990463257},{"id":"https://openalex.org/C66024118","wikidata":"https://www.wikidata.org/wiki/Q1122506","display_name":"Computational model","level":2,"score":0.28769999742507935},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.28360000252723694},{"id":"https://openalex.org/C141218545","wikidata":"https://www.wikidata.org/wiki/Q7521336","display_name":"Simulation language","level":2,"score":0.2741999924182892},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.271699994802475},{"id":"https://openalex.org/C23224414","wikidata":"https://www.wikidata.org/wiki/Q176769","display_name":"Hidden Markov model","level":2,"score":0.2502000033855438}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3746252.3761011","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3746252.3761011","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 34th ACM International Conference on Information and Knowledge Management","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.1145/3746252.3761011","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3746252.3761011","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 34th ACM International Conference on Information and Knowledge Management","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":5,"referenced_works":["https://openalex.org/W4382468457","https://openalex.org/W4385571988","https://openalex.org/W4396827130","https://openalex.org/W4402671545","https://openalex.org/W4409348217"],"related_works":[],"abstract_inverted_index":{"Enabled":[0],"by":[1],"large-scale":[2],"text":[3],"corpora":[4],"with":[5,189,198],"huge":[6],"parameters,":[7],"pre-trained":[8],"language":[9,68,156,166],"models":[10],"operate":[11],"as":[12,83],"multi-task":[13],"experts":[14,65],"using":[15,117],"a":[16,56,67,80,84,94,99,105,152,155,173,185,190,256],"single":[17],"model":[18,69,126,157],"architecture.":[19],"However,":[20],"recent":[21],"studies":[22,213],"have":[23],"revealed":[24],"that":[25,37,59,88,180,205],"certain":[26],"neurons":[27],"play":[28],"disproportionately":[29],"important":[30],"roles":[31],"in":[32],"solving":[33],"specific":[34,95],"tasks,":[35],"suggesting":[36],"task-relevant":[38,141],"substructures":[39],"can":[40],"be":[41],"isolated":[42],"and":[43,62,97,121,127,136,230,248],"selectively":[44],"activated":[45],"for":[46,129,239,259],"each":[47],"task.":[48,132],"Therefore,":[49],"we":[50,178,221],"introduce":[51],"Decomposition":[52],"of":[53,86,93,217,225],"Experts":[54],"(DoE),":[55],"novel":[57],"framework":[58,244],"dynamically":[60],"identifies":[61,208],"activates":[63],"task-specific":[64,261],"within":[66],"to":[70,184,251],"reduce":[71],"inference":[72,116,187,234],"cost":[73],"without":[74,194],"sacrificing":[75],"accuracy.":[76,196],"We":[77,150],"first":[78],"define":[79],"task":[81,96,112,148,171,200,209],"expert":[82,201],"set":[85],"parameters":[87],"significantly":[89],"influence":[90],"the":[91,110,118,124,130,215,223],"performance":[92],"propose":[98],"four-step":[100],"unplug-and-play":[101],"process:":[102],"(1)":[103],"receiving":[104],"user":[106,159],"request,":[107],"(2)":[108],"identifying":[109],"corresponding":[111],"expert,":[113],"(3)":[114],"performing":[115],"expert-localized":[119],"model,":[120],"(4)":[122],"restoring":[123],"original":[125],"waiting":[128],"next":[131],"Using":[133],"attribution":[134],"methods":[135,203],"prompt":[137],"tuning,":[138],"DoE":[139,181,206],"isolates":[140],"neurons,":[142],"minimizing":[143],"computational":[144],"overhead":[145],"while":[146,211],"maintaining":[147],"performance.":[149],"assume":[151],"setting":[153],"where":[154],"receives":[158],"requests":[160],"from":[161],"five":[162],"widely":[163],"used":[164],"natural":[165],"understanding":[167],"benchmarks,":[168],"processing":[169],"one":[170],"at":[172],"time.":[174],"In":[175],"this":[176],"setup,":[177],"demonstrate":[179],"achieves":[182],"up":[183],"x1.73":[186],"speed-up":[188],"65%":[191],"pruning":[192],"rate,":[193],"compromising":[195],"Comparisons":[197],"various":[199],"localization":[202],"reveal":[204],"effectively":[207],"experts,":[210],"ablation":[212],"validate":[214],"importance":[216],"its":[218],"components.":[219],"Additionally,":[220],"analyze":[222],"effects":[224],"batch":[226],"size,":[227],"token":[228],"count,":[229],"layer":[231],"types":[232],"on":[233],"speed-up,":[235],"providing":[236],"practical":[237,247],"insights":[238],"adopting":[240],"DoE.":[241],"The":[242],"proposed":[243],"is":[245],"both":[246],"scalable,":[249],"applicable":[250],"any":[252],"transformer-based":[253],"architecture,":[254],"offering":[255],"robust":[257],"solution":[258],"efficient":[260],"inference.":[262]},"counts_by_year":[],"updated_date":"2025-11-08T23:25:12.792448","created_date":"2025-11-08T00:00:00"}