{"id":"https://openalex.org/W4405095667","doi":"https://doi.org/10.1109/iccv51701.2025.00353","title":"p-MoD: Building Mixture-of-Depths MLLMs via Progressive Ratio Decay","display_name":"p-MoD: Building Mixture-of-Depths MLLMs via Progressive Ratio Decay","publication_year":2025,"publication_date":"2025-10-19","ids":{"openalex":"https://openalex.org/W4405095667","doi":"https://doi.org/10.1109/iccv51701.2025.00353"},"language":"en","primary_location":{"id":"doi:10.1109/iccv51701.2025.00353","is_oa":false,"landing_page_url":"https://doi.org/10.1109/iccv51701.2025.00353","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 IEEE/CVF International Conference on Computer Vision (ICCV)","raw_type":"proceedings-article"},"type":"preprint","indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2412.04449","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5100433105","display_name":"Jun Zhang","orcid":"https://orcid.org/0000-0002-0856-5693"},"institutions":[{"id":"https://openalex.org/I881766915","display_name":"Nanjing University","ror":"https://ror.org/01rxvg760","country_code":"CN","type":"education","lineage":["https://openalex.org/I881766915"]}],"countries":["CN"],"is_corresponding":true,"raw_author_name":"Jun Zhang","raw_affiliation_strings":["Nanjing University,State Key Laboratory for Novel Software Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Nanjing University,State Key Laboratory for Novel Software Technology","institution_ids":["https://openalex.org/I881766915"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5081405301","display_name":"Da Meng","orcid":"https://orcid.org/0000-0003-4451-4049"},"institutions":[{"id":"https://openalex.org/I881766915","display_name":"Nanjing University","ror":"https://ror.org/01rxvg760","country_code":"CN","type":"education","lineage":["https://openalex.org/I881766915"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Desen Meng","raw_affiliation_strings":["Nanjing University,State Key Laboratory for Novel Software Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Nanjing University,State Key Laboratory for Novel Software Technology","institution_ids":["https://openalex.org/I881766915"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5012916132","display_name":"Ji Qi","orcid":"https://orcid.org/0000-0001-7135-0524"},"institutions":[{"id":"https://openalex.org/I180662265","display_name":"China Mobile (China)","ror":"https://ror.org/05gftfe97","country_code":"CN","type":"company","lineage":["https://openalex.org/I180662265"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zhengming Zhang","raw_affiliation_strings":["China Mobile Research Institute"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"China Mobile Research Institute","institution_ids":["https://openalex.org/I180662265"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Zhenpeng Huang","orcid":null},"institutions":[{"id":"https://openalex.org/I881766915","display_name":"Nanjing University","ror":"https://ror.org/01rxvg760","country_code":"CN","type":"education","lineage":["https://openalex.org/I881766915"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zhenpeng Huang","raw_affiliation_strings":["Nanjing University,State Key Laboratory for Novel Software Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Nanjing University,State Key Laboratory for Novel Software Technology","institution_ids":["https://openalex.org/I881766915"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101478111","display_name":"Tao Wu","orcid":"https://orcid.org/0000-0003-2171-0389"},"institutions":[{"id":"https://openalex.org/I881766915","display_name":"Nanjing University","ror":"https://ror.org/01rxvg760","country_code":"CN","type":"education","lineage":["https://openalex.org/I881766915"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Tao Wu","raw_affiliation_strings":["Nanjing University,State Key Laboratory for Novel Software Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Nanjing University,State Key Laboratory for Novel Software Technology","institution_ids":["https://openalex.org/I881766915"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5115695199","display_name":"Limin Wang","orcid":"https://orcid.org/0000-0002-8849-9621"},"institutions":[{"id":"https://openalex.org/I881766915","display_name":"Nanjing University","ror":"https://ror.org/01rxvg760","country_code":"CN","type":"education","lineage":["https://openalex.org/I881766915"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Limin Wang","raw_affiliation_strings":["Nanjing University,State Key Laboratory for Novel Software Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Nanjing University,State Key Laboratory for Novel Software Technology","institution_ids":["https://openalex.org/I881766915"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":6,"corresponding_author_ids":["https://openalex.org/A5100433105"],"corresponding_institution_ids":["https://openalex.org/I881766915"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.0004482,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"3705","last_page":"3715"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12808","display_name":"Ferroelectric and Negative Capacitance Devices","score":0.9973999857902527,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T12808","display_name":"Ferroelectric and Negative Capacitance Devices","score":0.9973999857902527,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10502","display_name":"Advanced Memory and Neural Computing","score":0.9818000197410583,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10207","display_name":"Advanced biosensing and bioanalysis techniques","score":0.9803000092506409,"subfield":{"id":"https://openalex.org/subfields/1312","display_name":"Molecular Biology"},"field":{"id":"https://openalex.org/fields/13","display_name":"Biochemistry, Genetics and Molecular Biology"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/mod","display_name":"Mod","score":0.8636647462844849},{"id":"https://openalex.org/keywords/mathematics","display_name":"Mathematics","score":0.37372875213623047},{"id":"https://openalex.org/keywords/environmental-science","display_name":"Environmental science","score":0.34256917238235474},{"id":"https://openalex.org/keywords/physics","display_name":"Physics","score":0.3230936825275421},{"id":"https://openalex.org/keywords/combinatorics","display_name":"Combinatorics","score":0.21144068241119385}],"concepts":[{"id":"https://openalex.org/C29231244","wikidata":"https://www.wikidata.org/wiki/Q865493","display_name":"Mod","level":2,"score":0.8636647462844849},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.37372875213623047},{"id":"https://openalex.org/C39432304","wikidata":"https://www.wikidata.org/wiki/Q188847","display_name":"Environmental science","level":0,"score":0.34256917238235474},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.3230936825275421},{"id":"https://openalex.org/C114614502","wikidata":"https://www.wikidata.org/wiki/Q76592","display_name":"Combinatorics","level":1,"score":0.21144068241119385}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.1109/iccv51701.2025.00353","is_oa":false,"landing_page_url":"https://doi.org/10.1109/iccv51701.2025.00353","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 IEEE/CVF International Conference on Computer Vision (ICCV)","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2412.04449","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2412.04449","pdf_url":"https://arxiv.org/pdf/2412.04449","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"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":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2412.04449","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2412.04449","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"article"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2412.04449","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2412.04449","pdf_url":"https://arxiv.org/pdf/2412.04449","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"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":null,"raw_type":"text"},"sustainable_development_goals":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320324852","display_name":"Nanjing University","ror":"https://ror.org/01rxvg760"}],"has_content":{"pdf":true,"grobid_xml":false},"content_urls":{"pdf":"https://content.openalex.org/works/W4405095667.pdf"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2935759653","https://openalex.org/W3105167352","https://openalex.org/W54078636","https://openalex.org/W2954470139","https://openalex.org/W1998970406","https://openalex.org/W1501425562","https://openalex.org/W2902782467","https://openalex.org/W3084825885","https://openalex.org/W2298861036"],"abstract_inverted_index":{"Despite":[0],"the":[1,13,51,59,65,92,107,147,164,170,194],"remarkable":[2],"performance":[3,173,195],"of":[4,45,54,94,166,174,196],"multimodal":[5],"large":[6],"language":[7],"models":[8,182],"(MLLMs)":[9],"across":[10,183],"diverse":[11],"tasks,":[12],"substantial":[14],"training":[15,35,95,103],"and":[16,36,96,118,135,172,204,211],"inference":[17,37,97],"costs":[18,38],"impede":[19],"their":[20],"advancement.":[21],"In":[22],"this":[23],"paper,":[24],"we":[25,63,105,124],"propose":[26],"p-MoD,":[27],"an":[28],"efficient":[29],"MLLM":[30],"architecture":[31],"that":[32,126,187],"significantly":[33,168],"reduces":[34,146],"while":[39,79,199],"maintaining":[40],"model":[41,189],"performance.":[42],"The":[43],"majority":[44],"computation":[46],"in":[47,132],"MLLMs":[48,87],"stems":[49],"from":[50],"overwhelming":[52],"volume":[53],"vision":[55,75,127],"tokens":[56,76,128],"processed":[57],"by":[58,152],"transformer-based":[60],"LLM.":[61],"Accordingly,":[62],"leverage":[64],"Mixture-of-Depths":[66],"(MoD)":[67],"mechanism,":[68],"where":[69],"each":[70],"LLM":[71],"layer":[72,151],"selects":[73],"essential":[74],"to":[77],"process":[78],"skipping":[80],"redundant":[81],"ones.":[82],"However,":[83],"integrating":[84],"MoD":[85,108],"into":[86],"is":[88],"non-trivial.":[89],"To":[90],"address":[91],"challenges":[93],"stability":[98],"as":[99,101],"well":[100],"limited":[102],"data,":[104],"adapt":[106],"module":[109],"with":[110],"two":[111,180],"novel":[112],"designs:":[113],"tanh-gated":[114],"weight":[115],"normalization":[116],"(TanhNorm)":[117],"symmetric":[119],"token":[120,148],"reweighting":[121],"(STRing).":[122],"Moreover,":[123],"observe":[125],"exhibit":[129],"higher":[130],"redundancy":[131],"deeper":[133],"layers":[134],"thus":[136],"design":[137,161],"a":[138,155],"progressive":[139],"ratio":[140,150],"decay":[141],"(PRD)":[142],"strategy,":[143],"which":[144],"gradually":[145],"retention":[149],"layer,":[153],"employing":[154],"shifted":[156],"cosine":[157],"schedule.":[158],"This":[159],"crucial":[160],"fully":[162],"unleashes":[163],"potential":[165],"MoD,":[167],"boosting":[169],"efficiency":[171],"our":[175,188],"models.":[176],"Extensive":[177],"experiments":[178],"on":[179],"baseline":[181],"15":[184],"benchmarks":[185],"show":[186],"matches":[190],"or":[191],"even":[192],"surpasses":[193],"corresponding":[197],"baselines,":[198],"requiring":[200],"only":[201],"55.6%":[202],"TFLOPs":[203],"53.7%":[205],"KV":[206],"cache":[207],"storage":[208],"during":[209,215],"inference,":[210],"77.7%":[212],"GPU":[213],"hours":[214],"training.":[216]},"counts_by_year":[],"updated_date":"2026-05-06T06:03:25.996018","created_date":"2025-10-10T00:00:00"}