{"id":"https://openalex.org/W7124471359","doi":"https://doi.org/10.1109/tmm.2026.3654458","title":"MoE-LLaVA: Mixture of Experts for Large Vision-Language Models","display_name":"MoE-LLaVA: Mixture of Experts for Large Vision-Language Models","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7124471359","doi":"https://doi.org/10.1109/tmm.2026.3654458"},"language":null,"primary_location":{"id":"doi:10.1109/tmm.2026.3654458","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tmm.2026.3654458","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":"https://openalex.org/A5123180841","display_name":"Bin Lin","orcid":null},"institutions":[{"id":"https://openalex.org/I180726961","display_name":"Shenzhen University","ror":"https://ror.org/01vy4gh70","country_code":"CN","type":"education","lineage":["https://openalex.org/I180726961"]},{"id":"https://openalex.org/I4210128628","display_name":"Peking University Shenzhen Hospital","ror":"https://ror.org/03kkjyb15","country_code":"CN","type":"healthcare","lineage":["https://openalex.org/I4210128628"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Bin Lin","raw_affiliation_strings":["College of Information Engineering, Peking Univercity Shenzhen Graduate School, Shenzhen, China"],"raw_orcid":"https://orcid.org/0009-0003-4805-9730","affiliations":[{"raw_affiliation_string":"College of Information Engineering, Peking Univercity Shenzhen Graduate School, Shenzhen, China","institution_ids":["https://openalex.org/I180726961","https://openalex.org/I4210128628"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123219491","display_name":"Zhenyu Tang","orcid":null},"institutions":[{"id":"https://openalex.org/I180726961","display_name":"Shenzhen University","ror":"https://ror.org/01vy4gh70","country_code":"CN","type":"education","lineage":["https://openalex.org/I180726961"]},{"id":"https://openalex.org/I4210128628","display_name":"Peking University Shenzhen Hospital","ror":"https://ror.org/03kkjyb15","country_code":"CN","type":"healthcare","lineage":["https://openalex.org/I4210128628"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zhenyu Tang","raw_affiliation_strings":["College of Information Engineering, Peking Univercity Shenzhen Graduate School, Shenzhen, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"College of Information Engineering, Peking Univercity Shenzhen Graduate School, Shenzhen, China","institution_ids":["https://openalex.org/I180726961","https://openalex.org/I4210128628"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123175859","display_name":"Yang Ye","orcid":null},"institutions":[{"id":"https://openalex.org/I180726961","display_name":"Shenzhen University","ror":"https://ror.org/01vy4gh70","country_code":"CN","type":"education","lineage":["https://openalex.org/I180726961"]},{"id":"https://openalex.org/I4210128628","display_name":"Peking University Shenzhen Hospital","ror":"https://ror.org/03kkjyb15","country_code":"CN","type":"healthcare","lineage":["https://openalex.org/I4210128628"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yang Ye","raw_affiliation_strings":["College of Information Engineering, Peking Univercity Shenzhen Graduate School, Shenzhen, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"College of Information Engineering, Peking Univercity Shenzhen Graduate School, Shenzhen, China","institution_ids":["https://openalex.org/I180726961","https://openalex.org/I4210128628"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123201124","display_name":"Jinfa Huang","orcid":null},"institutions":[{"id":"https://openalex.org/I5388228","display_name":"University of Rochester","ror":"https://ror.org/022kthw22","country_code":"US","type":"education","lineage":["https://openalex.org/I5388228"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Jinfa Huang","raw_affiliation_strings":["Department of Computer Science, University of Rochester Rochester, NY, USA"],"raw_orcid":"https://orcid.org/0000-0002-0081-4106","affiliations":[{"raw_affiliation_string":"Department of Computer Science, University of Rochester Rochester, NY, USA","institution_ids":["https://openalex.org/I5388228"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123208227","display_name":"Junwu Zhang","orcid":null},"institutions":[{"id":"https://openalex.org/I180726961","display_name":"Shenzhen University","ror":"https://ror.org/01vy4gh70","country_code":"CN","type":"education","lineage":["https://openalex.org/I180726961"]},{"id":"https://openalex.org/I4210128628","display_name":"Peking University Shenzhen Hospital","ror":"https://ror.org/03kkjyb15","country_code":"CN","type":"healthcare","lineage":["https://openalex.org/I4210128628"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Junwu Zhang","raw_affiliation_strings":["College of Information Engineering, Peking Univercity Shenzhen Graduate School, Shenzhen, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"College of Information Engineering, Peking Univercity Shenzhen Graduate School, Shenzhen, China","institution_ids":["https://openalex.org/I180726961","https://openalex.org/I4210128628"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5057725243","display_name":"Yatian Pang","orcid":"https://orcid.org/0000-0002-9714-9068"},"institutions":[{"id":"https://openalex.org/I180726961","display_name":"Shenzhen University","ror":"https://ror.org/01vy4gh70","country_code":"CN","type":"education","lineage":["https://openalex.org/I180726961"]},{"id":"https://openalex.org/I4210128628","display_name":"Peking University Shenzhen Hospital","ror":"https://ror.org/03kkjyb15","country_code":"CN","type":"healthcare","lineage":["https://openalex.org/I4210128628"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yatian Pang","raw_affiliation_strings":["College of Information Engineering, Peking Univercity Shenzhen Graduate School, Shenzhen, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"College of Information Engineering, Peking Univercity Shenzhen Graduate School, Shenzhen, China","institution_ids":["https://openalex.org/I180726961","https://openalex.org/I4210128628"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123185377","display_name":"Peng Jin","orcid":null},"institutions":[{"id":"https://openalex.org/I180726961","display_name":"Shenzhen University","ror":"https://ror.org/01vy4gh70","country_code":"CN","type":"education","lineage":["https://openalex.org/I180726961"]},{"id":"https://openalex.org/I4210128628","display_name":"Peking University Shenzhen Hospital","ror":"https://ror.org/03kkjyb15","country_code":"CN","type":"healthcare","lineage":["https://openalex.org/I4210128628"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Peng Jin","raw_affiliation_strings":["College of Information Engineering, Peking Univercity Shenzhen Graduate School, Shenzhen, China"],"raw_orcid":"https://orcid.org/0000-0001-9287-6410","affiliations":[{"raw_affiliation_string":"College of Information Engineering, Peking Univercity Shenzhen Graduate School, Shenzhen, China","institution_ids":["https://openalex.org/I180726961","https://openalex.org/I4210128628"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5058644424","display_name":"Munan Ning","orcid":"https://orcid.org/0009-0005-3418-085X"},"institutions":[{"id":"https://openalex.org/I180726961","display_name":"Shenzhen University","ror":"https://ror.org/01vy4gh70","country_code":"CN","type":"education","lineage":["https://openalex.org/I180726961"]},{"id":"https://openalex.org/I4210128628","display_name":"Peking University Shenzhen Hospital","ror":"https://ror.org/03kkjyb15","country_code":"CN","type":"healthcare","lineage":["https://openalex.org/I4210128628"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Munan Ning","raw_affiliation_strings":["College of Information Engineering, Peking Univercity Shenzhen Graduate School, Shenzhen, China"],"raw_orcid":"https://orcid.org/0009-0005-3418-085X","affiliations":[{"raw_affiliation_string":"College of Information Engineering, Peking Univercity Shenzhen Graduate School, Shenzhen, China","institution_ids":["https://openalex.org/I180726961","https://openalex.org/I4210128628"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123208294","display_name":"Jiebo Luo","orcid":null},"institutions":[{"id":"https://openalex.org/I5388228","display_name":"University of Rochester","ror":"https://ror.org/022kthw22","country_code":"US","type":"education","lineage":["https://openalex.org/I5388228"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Jiebo Luo","raw_affiliation_strings":["Department of Computer Science, University of Rochester Rochester, NY, USA"],"raw_orcid":"https://orcid.org/0000-0002-4516-9729","affiliations":[{"raw_affiliation_string":"Department of Computer Science, University of Rochester Rochester, NY, USA","institution_ids":["https://openalex.org/I5388228"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5123239152","display_name":"Li Yuan","orcid":null},"institutions":[{"id":"https://openalex.org/I180726961","display_name":"Shenzhen University","ror":"https://ror.org/01vy4gh70","country_code":"CN","type":"education","lineage":["https://openalex.org/I180726961"]},{"id":"https://openalex.org/I4210128628","display_name":"Peking University Shenzhen Hospital","ror":"https://ror.org/03kkjyb15","country_code":"CN","type":"healthcare","lineage":["https://openalex.org/I4210128628"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Li Yuan","raw_affiliation_strings":["College of Information Engineering, Peking Univercity Shenzhen Graduate School, Shenzhen, China"],"raw_orcid":"https://orcid.org/0000-0002-2120-5588","affiliations":[{"raw_affiliation_string":"College of Information Engineering, Peking Univercity Shenzhen Graduate School, Shenzhen, China","institution_ids":["https://openalex.org/I180726961","https://openalex.org/I4210128628"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":3,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":98.1143,"has_fulltext":false,"cited_by_count":15,"citation_normalized_percentile":{"value":0.99937254,"is_in_top_1_percent":true,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":95,"max":100},"biblio":{"volume":"28","issue":null,"first_page":"4408","last_page":"4419"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.8109999895095825,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.8109999895095825,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.08619999885559082,"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":0.023600000888109207,"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/initialization","display_name":"Initialization","score":0.673799991607666},{"id":"https://openalex.org/keywords/sparse-matrix","display_name":"Sparse matrix","score":0.5907999873161316},{"id":"https://openalex.org/keywords/sparse-approximation","display_name":"Sparse approximation","score":0.567799985408783},{"id":"https://openalex.org/keywords/feature","display_name":"Feature (linguistics)","score":0.42010000348091125},{"id":"https://openalex.org/keywords/code","display_name":"Code (set theory)","score":0.3982999920845032},{"id":"https://openalex.org/keywords/baseline","display_name":"Baseline (sea)","score":0.3788999915122986},{"id":"https://openalex.org/keywords/lossless-compression","display_name":"Lossless compression","score":0.3725000023841858}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.821399986743927},{"id":"https://openalex.org/C114466953","wikidata":"https://www.wikidata.org/wiki/Q6034165","display_name":"Initialization","level":2,"score":0.673799991607666},{"id":"https://openalex.org/C56372850","wikidata":"https://www.wikidata.org/wiki/Q1050404","display_name":"Sparse matrix","level":3,"score":0.5907999873161316},{"id":"https://openalex.org/C124066611","wikidata":"https://www.wikidata.org/wiki/Q28684319","display_name":"Sparse approximation","level":2,"score":0.567799985408783},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5184000134468079},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4871000051498413},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.42010000348091125},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.3982999920845032},{"id":"https://openalex.org/C12725497","wikidata":"https://www.wikidata.org/wiki/Q810247","display_name":"Baseline (sea)","level":2,"score":0.3788999915122986},{"id":"https://openalex.org/C81081738","wikidata":"https://www.wikidata.org/wiki/Q55542","display_name":"Lossless compression","level":3,"score":0.3725000023841858},{"id":"https://openalex.org/C77637269","wikidata":"https://www.wikidata.org/wiki/Q7002051","display_name":"Neural coding","level":2,"score":0.35280001163482666},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.3249000012874603},{"id":"https://openalex.org/C199519371","wikidata":"https://www.wikidata.org/wiki/Q942695","display_name":"Source lines of code","level":3,"score":0.30809998512268066},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.30649998784065247},{"id":"https://openalex.org/C2777027219","wikidata":"https://www.wikidata.org/wiki/Q1284190","display_name":"Constant (computer programming)","level":2,"score":0.3043000102043152},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.2994999885559082},{"id":"https://openalex.org/C120936955","wikidata":"https://www.wikidata.org/wiki/Q2155640","display_name":"Empirical research","level":2,"score":0.29919999837875366},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.2621999979019165}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/tmm.2026.3654458","is_oa":false,"landing_page_url":"https://doi.org/10.1109/tmm.2026.3654458","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":[],"awards":[{"id":"https://openalex.org/G1947953275","display_name":null,"funder_award_id":"62425101","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"},{"id":"https://openalex.org/G5151253548","display_name":null,"funder_award_id":"62332002","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":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Recently,":[0],"remarkable":[1],"progress":[2],"has":[3],"been":[4],"made":[5],"in":[6],"scaling":[7],"up":[8],"Large":[9,37],"Language":[10],"Models":[11],"(LLMs)":[12],"through":[13,168],"the":[14,17,28,84,88,93,99,102,106,162,173,228],"use":[15],"of":[16,126,188],"sparse":[18,36,53,120,145,155,220,229],"Mixture-of-Expert":[19],"(MoE)":[20],"layers":[21],"without":[22],"significantly":[23],"increasing":[24],"computational":[25,131],"cost.":[26,132],"However,":[27],"transition":[29],"from":[30,105],"a":[31,35,52,63,77,119,123,144,153,217],"pre-trained":[32],"LLM":[33,50,94],"to":[34,51,57,91,95],"Vision-Language":[38],"Model":[39],"(LVLM)":[40],"with":[41,122,184,196,206],"MoE":[42],"remains":[43],"an":[44,49,96,185],"open":[45],"challenge.":[46],"Directly":[47],"fine-tuning":[48],"LVLM":[54,146,156],"often":[55],"leads":[56],"training":[58,81],"collapse,":[59],"characterized":[60],"by":[61],"(1)":[62],"large":[64],"modality":[65],"feature":[66],"distribution":[67],"gap":[68],"and":[69,142,222],"(2)":[70],"expert":[71,115],"load":[72],"imbalance.":[73],"This":[74],"paper":[75],"proposes":[76],"three-stage":[78],"decoupled":[79],"weight":[80],"process.":[82],"In":[83,98],"first":[85],"two":[86],"stages,":[87],"model":[89,121],"learns":[90],"adapt":[92],"LVLM.":[97],"third":[100],"stage,":[101],"FFN":[103],"weights":[104],"second":[107],"stage":[108],"are":[109],"used":[110],"as":[111],"lossless":[112],"initialization":[113],"for":[114,219,226],"weights,":[116],"effectively":[117],"constructing":[118],"vast":[124],"number":[125],"parameters":[127],"while":[128],"maintaining":[129],"constant":[130],"Through":[133],"extensive":[134],"ablation":[135],"experiments,":[136],"we":[137],"derive":[138],"three":[139],"empirical":[140,224],"guidelines":[141,225],"propose":[143],"termed":[147],"<bold":[148],"xmlns:mml=\"http://www.w3.org/1998/Math/MathML\"":[149,164,237],"xmlns:xlink=\"http://www.w3.org/1999/xlink\">MoE-LLaVA</b>.":[150],"MoE-LLaVA":[151,181,202],"is":[152,233],"MoE-based":[154],"architecture,":[157],"which":[158],"uniquely":[159],"activates":[160],"only":[161,197],"top-<inline-formula":[163],"xmlns:xlink=\"http://www.w3.org/1999/xlink\"><tex-math":[165],"notation=\"LaTeX\">$k$</tex-math></inline-formula>":[166],"experts":[167,175],"routers":[169],"during":[170],"deployment,":[171],"keeping":[172],"remaining":[174],"inactive.":[176],"Extensive":[177],"experiments":[178],"demonstrate":[179],"that":[180],"outperforms":[182],"LLaVA-1.5-7B":[183],"average":[186],"improvement":[187],"4.6":[189],"across":[190],"nine":[191],"visual":[192],"understanding":[193],"benchmarks.":[194],"Notably,":[195],"2.2B":[198],"active":[199],"parameters,":[200],"our":[201],"shows":[203],"comparable":[204],"result":[205],"LLaVA-1.5-13B":[207],"(87.0":[208],"vs.":[209],"85.9)":[210],"on":[211],"POPE":[212],"benchmark.":[213],"Our":[214,231],"work":[215],"establishes":[216],"baseline":[218],"LVLMs":[221],"provides":[223],"exploring":[227],"LVLMs.":[230],"code":[232],"available":[234],"at:":[235],"<uri":[236],"xmlns:xlink=\"http://www.w3.org/1999/xlink\">https://github.com/PKU-YuanGroup/MoE-LLaVA</uri>.":[238]},"counts_by_year":[{"year":2026,"cited_by_count":13},{"year":2025,"cited_by_count":2}],"updated_date":"2026-07-17T09:13:05.818461","created_date":"2026-01-17T00:00:00"}
