{"id":"https://openalex.org/W7138016973","doi":"https://doi.org/10.1609/aaai.v40i36.40319","title":"MHA2MLA-VLM: Enabling DeepSeek\u2019s Economical Multi-Head Latent Attention Across Vision-Language Models","display_name":"MHA2MLA-VLM: Enabling DeepSeek\u2019s Economical Multi-Head Latent Attention Across Vision-Language Models","publication_year":2026,"publication_date":"2026-03-14","ids":{"openalex":"https://openalex.org/W7138016973","doi":"https://doi.org/10.1609/aaai.v40i36.40319"},"language":null,"primary_location":{"id":"doi:10.1609/aaai.v40i36.40319","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i36.40319","pdf_url":null,"source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"diamond","oa_url":"https://doi.org/10.1609/aaai.v40i36.40319","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5101588749","display_name":"Xiaoran Fan","orcid":"https://orcid.org/0000-0002-6368-9250"},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Xiaoran Fan","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129712243","display_name":"Zhichao Sun","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhichao Sun","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129669213","display_name":"Tao Ji","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tao Ji","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123348893","display_name":"Lixing Shen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lixing Shen","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5129700592","display_name":"Tao Gui","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tao Gui","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":5,"corresponding_author_ids":["https://openalex.org/A5101588749"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.29903057,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":"40","issue":"36","first_page":"30638","last_page":"30646"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.9656000137329102,"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":0.9656000137329102,"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":0.011800000444054604,"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":0.005799999926239252,"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/cache","display_name":"Cache","score":0.6503000259399414},{"id":"https://openalex.org/keywords/masking","display_name":"Masking (illustration)","score":0.5767999887466431},{"id":"https://openalex.org/keywords/adaptation","display_name":"Adaptation (eye)","score":0.5400999784469604},{"id":"https://openalex.org/keywords/online-model","display_name":"Online model","score":0.36410000920295715},{"id":"https://openalex.org/keywords/core","display_name":"Core (optical fiber)","score":0.33899998664855957}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8210999965667725},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.6503000259399414},{"id":"https://openalex.org/C2777402240","wikidata":"https://www.wikidata.org/wiki/Q6783436","display_name":"Masking (illustration)","level":2,"score":0.5767999887466431},{"id":"https://openalex.org/C139807058","wikidata":"https://www.wikidata.org/wiki/Q352374","display_name":"Adaptation (eye)","level":2,"score":0.5400999784469604},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5091000199317932},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4068000018596649},{"id":"https://openalex.org/C2777851325","wikidata":"https://www.wikidata.org/wiki/Q7094102","display_name":"Online model","level":2,"score":0.36410000920295715},{"id":"https://openalex.org/C2164484","wikidata":"https://www.wikidata.org/wiki/Q5170150","display_name":"Core (optical fiber)","level":2,"score":0.33899998664855957},{"id":"https://openalex.org/C189783530","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"CPU cache","level":3,"score":0.289900004863739},{"id":"https://openalex.org/C123657996","wikidata":"https://www.wikidata.org/wiki/Q12271","display_name":"Architecture","level":2,"score":0.27970001101493835},{"id":"https://openalex.org/C78766204","wikidata":"https://www.wikidata.org/wiki/Q555032","display_name":"Multi-core processor","level":2,"score":0.2502000033855438}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1609/aaai.v40i36.40319","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i36.40319","pdf_url":null,"source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"}],"best_oa_location":{"id":"doi:10.1609/aaai.v40i36.40319","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i36.40319","pdf_url":null,"source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"As":[0],"vision-language":[1],"models":[2],"(VLMs)":[3],"tackle":[4],"increasingly":[5],"complex":[6],"and":[7,20,39,63,87,95,107,120,157],"multimodal":[8,88],"tasks,":[9],"the":[10,36,46,105],"rapid":[11],"growth":[12],"of":[13],"Key-Value":[14],"(KV)":[15],"cache":[16,38,155],"imposes":[17],"significant":[18],"memory":[19],"computational":[21],"bottlenecks":[22],"during":[23],"inference.":[24],"While":[25],"Multi-Head":[26],"Latent":[27],"Attention":[28],"(MLA)":[29],"offers":[30],"an":[31],"effective":[32],"means":[33],"to":[34,45,70,116],"compress":[35],"KV":[37,109,154,161],"accelerate":[40],"inference,":[41],"adapting":[42],"existing":[43],"VLMs":[44,69,140],"MLA":[47],"architecture":[48],"without":[49],"costly":[50],"pretraining":[51],"remains":[52],"largely":[53],"unexplored.":[54],"In":[55],"this":[56],"work,":[57],"we":[58,112],"present":[59],"\\textbf{MHA2MLA-VLM},":[60],"a":[61,79,97],"parameter-efficient":[62,114],"multimodal-aware":[64],"framework":[65],"for":[66],"converting":[67],"off-the-shelf":[68],"MLA.":[71],"Our":[72],"approach":[73],"features":[74],"two":[75],"core":[76],"techniques:":[77],"(1)":[78],"modality-adaptive":[80],"partial-RoPE":[81],"strategy":[82],"that":[83,102,122,142],"supports":[84],"both":[85],"traditional":[86],"settings":[89],"by":[90],"selectively":[91],"masking":[92],"nonessential":[93],"dimensions,":[94],"(2)":[96],"modality-decoupled":[98],"low-rank":[99],"approximation":[100],"method":[101],"independently":[103],"compresses":[104],"visual":[106],"textual":[108],"spaces.":[110],"Furthermore,":[111],"introduce":[113],"fine-tuning":[115],"minimize":[117],"adaptation":[118],"cost":[119],"demonstrate":[121],"minimizing":[123],"output":[124],"activation":[125],"error,":[126],"rather":[127],"than":[128],"parameter":[129],"distance,":[130],"substantially":[131],"reduces":[132,153],"performance":[133,147],"loss.":[134],"Extensive":[135],"experiments":[136],"on":[137],"three":[138],"representative":[139],"show":[141],"MHA2MLA-VLM":[143],"restores":[144],"original":[145],"model":[146],"with":[148,160],"minimal":[149],"supervised":[150],"data,":[151],"significantly":[152],"footprint,":[156],"integrates":[158],"seamlessly":[159],"quantization.":[162]},"counts_by_year":[],"updated_date":"2026-03-18T06:31:55.123368","created_date":"2026-03-18T00:00:00"}