{"id":"https://openalex.org/W7168187581","doi":"https://doi.org/10.48550/arxiv.2607.09029","title":"MOSAIC: Adaptive Inter-layer Composition for Efficient Heterogeneous Vision-Language Models","display_name":"MOSAIC: Adaptive Inter-layer Composition for Efficient Heterogeneous Vision-Language Models","publication_year":2026,"publication_date":"2026-07-10","ids":{"openalex":"https://openalex.org/W7168187581","doi":"https://doi.org/10.48550/arxiv.2607.09029"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2607.09029","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2607.09029","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2607.09029","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5052688308","display_name":"Yuncheng Yang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yang, Yuncheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5140540740","display_name":"Feiyang Ye","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ye, Feiyang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5041301168","display_name":"Shixian Luo","orcid":"https://orcid.org/0000-0003-1538-0950"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Luo, Shixian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123461994","display_name":"Yinna Zhu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhu, Yinna","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5010380116","display_name":"Lianlei Shan","orcid":"https://orcid.org/0000-0002-4648-8246"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shan, Lianlei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5016584164","display_name":"Weidong Zhao","orcid":"https://orcid.org/0000-0003-4144-7303"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao, Wangcai","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5065782115","display_name":"K Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Kuo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5140527671","display_name":"Yan Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Yan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5140561844","display_name":"Yong Wu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wu, Yong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5140534871","display_name":"Yan Xie","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xie, Yan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.8183000087738037,"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":0.8183000087738037,"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":0.05700000002980232,"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":0.031300000846385956,"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/inference","display_name":"Inference","score":0.5680999755859375},{"id":"https://openalex.org/keywords/oracle","display_name":"Oracle","score":0.5253000259399414},{"id":"https://openalex.org/keywords/interleaving","display_name":"Interleaving","score":0.4878000020980835},{"id":"https://openalex.org/keywords/bottleneck","display_name":"Bottleneck","score":0.4293000102043152},{"id":"https://openalex.org/keywords/latency","display_name":"Latency (audio)","score":0.4058000147342682},{"id":"https://openalex.org/keywords/homogeneous","display_name":"Homogeneous","score":0.3971000015735626},{"id":"https://openalex.org/keywords/linear-programming","display_name":"Linear programming","score":0.37209999561309814},{"id":"https://openalex.org/keywords/integer-programming","display_name":"Integer programming","score":0.3626999855041504},{"id":"https://openalex.org/keywords/process","display_name":"Process (computing)","score":0.35260000824928284}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7459999918937683},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5680999755859375},{"id":"https://openalex.org/C55166926","wikidata":"https://www.wikidata.org/wiki/Q2892946","display_name":"Oracle","level":2,"score":0.5253000259399414},{"id":"https://openalex.org/C28034677","wikidata":"https://www.wikidata.org/wiki/Q17092530","display_name":"Interleaving","level":2,"score":0.4878000020980835},{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.4293000102043152},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.4058000147342682},{"id":"https://openalex.org/C66882249","wikidata":"https://www.wikidata.org/wiki/Q169336","display_name":"Homogeneous","level":2,"score":0.3971000015735626},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.37369999289512634},{"id":"https://openalex.org/C41045048","wikidata":"https://www.wikidata.org/wiki/Q202843","display_name":"Linear programming","level":2,"score":0.37209999561309814},{"id":"https://openalex.org/C56086750","wikidata":"https://www.wikidata.org/wiki/Q6042592","display_name":"Integer programming","level":2,"score":0.3626999855041504},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.35260000824928284},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.34769999980926514},{"id":"https://openalex.org/C206588197","wikidata":"https://www.wikidata.org/wiki/Q846574","display_name":"Reuse","level":2,"score":0.32199999690055847},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.31690001487731934},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.30720001459121704},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.3027999997138977},{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.29589998722076416},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.2904999852180481},{"id":"https://openalex.org/C61326573","wikidata":"https://www.wikidata.org/wiki/Q1496376","display_name":"Gaussian process","level":3,"score":0.2897999882698059},{"id":"https://openalex.org/C147764199","wikidata":"https://www.wikidata.org/wiki/Q6865248","display_name":"Minification","level":2,"score":0.28679999709129333},{"id":"https://openalex.org/C41608201","wikidata":"https://www.wikidata.org/wiki/Q980509","display_name":"Embedding","level":2,"score":0.2865000069141388},{"id":"https://openalex.org/C204030448","wikidata":"https://www.wikidata.org/wiki/Q101017","display_name":"Distillation","level":2,"score":0.2824999988079071},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.2786000072956085},{"id":"https://openalex.org/C179799912","wikidata":"https://www.wikidata.org/wiki/Q205084","display_name":"Computational complexity theory","level":2,"score":0.2727999985218048},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.2696000039577484},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2646999955177307},{"id":"https://openalex.org/C93959086","wikidata":"https://www.wikidata.org/wiki/Q6888345","display_name":"Model selection","level":2,"score":0.2619999945163727},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.2596000134944916},{"id":"https://openalex.org/C173404611","wikidata":"https://www.wikidata.org/wiki/Q528588","display_name":"Constraint programming","level":3,"score":0.25839999318122864},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.25519999861717224},{"id":"https://openalex.org/C175291020","wikidata":"https://www.wikidata.org/wiki/Q1156822","display_name":"Offset (computer science)","level":2,"score":0.2529999911785126}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2607.09029","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2607.09029","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2607.09029","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2607.09029","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Vision-Language":[0],"Models":[1],"(VLMs)":[2],"have":[3],"achieved":[4],"success":[5],"using":[6],"homogeneous":[7,32,73],"Transformers":[8],"to":[9,48,50,136],"process":[10],"multimedia":[11],"data.":[12],"Recent":[13],"studies":[14],"show":[15],"that":[16,70,110,171],"heterogeneous":[17,77],"structures":[18],"interleaving":[19],"efficient":[20],"mechanisms,":[21],"like":[22],"linear":[23],"attention,":[24],"improve":[25],"both":[26],"performance":[27,113,121,176],"and":[28,46,86,153,199],"inference":[29,194],"latency":[30,117],"over":[31],"designs.":[33],"However,":[34],"these":[35],"efforts":[36],"rely":[37],"on":[38],"handcrafted":[39],"static":[40],"mixing":[41],"patterns,":[42],"which":[43],"are":[44],"sub-optimal":[45],"difficult":[47],"adapt":[49],"specific":[51],"hardware.":[52],"To":[53,119],"bridge":[54],"this":[55],"gap,":[56],"we":[57,126],"propose":[58],"Multi-Objective":[59],"Search":[60],"for":[61,150,158],"Adaptive":[62],"Inter-layer":[63],"Composition":[64],"(MOSAIC),":[65],"a":[66,89,98,128,142,147],"hardware-aware":[67],"search":[68,91],"method":[69,106],"automatically":[71],"transforms":[72],"models":[74],"into":[75],"optimized":[76],"architectures.":[78],"MOSAIC":[79,163],"integrates":[80],"diverse":[81],"efficiency":[82],"mechanisms--including":[83],"linear,":[84],"sparse,":[85],"low-rank":[87],"operators--into":[88],"unified":[90],"space.":[92],"By":[93],"formulating":[94],"the":[95,154,174,186],"selection":[96],"as":[97],"multi-objective":[99],"Mixed":[100],"Integer":[101],"Programming":[102],"(MIP)":[103],"problem,":[104],"our":[105],"identifies":[107],"optimal":[108],"configurations":[109],"maximize":[111],"downstream":[112],"under":[114],"strict":[115],"hardware":[116],"constraints.":[118],"mitigate":[120],"degradation":[122],"from":[123,167],"structural":[124],"transitions,":[125],"introduce":[127],"two-stage":[129],"parameter":[130],"recovery":[131],"process:":[132],"global":[133],"off-policy":[134],"distillation":[135,145],"stabilize":[137],"internal":[138],"representations,":[139],"followed":[140],"by":[141],"dual-teacher":[143],"on-policy":[144],"leveraging":[146],"235B":[148],"oracle":[149],"knowledge":[151],"expansion":[152],"original":[155,187],"4B":[156],"teacher":[157],"distributional":[159],"stability.":[160],"We":[161],"validate":[162],"through":[164],"MOSAIC-4B,":[165],"derived":[166],"Qwen3-VL-4B-Instruct.":[168],"Results":[169],"demonstrate":[170],"MOSAIC-4B":[172],"matches":[173],"baseline's":[175],"across":[177],"multiple":[178],"benchmarks":[179],"while":[180],"requiring":[181],"less":[182],"than":[183],"2%":[184],"of":[185],"training":[188],"cost.":[189],"Furthermore,":[190],"it":[191],"substantially":[192],"improves":[193],"efficiency,":[195],"achieving":[196],"1.76x":[197],"prefilling":[198],"2.54x":[200],"decoding":[201],"speedups.":[202]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-07-14T00:00:00"}