{"id":"https://openalex.org/W4412944615","doi":"https://doi.org/10.18653/v1/2025.findings-acl.1233","title":"RedundancyLens: Revealing and Exploiting Visual Token Processing Redundancy for Efficient Decoder-Only MLLMs","display_name":"RedundancyLens: Revealing and Exploiting Visual Token Processing Redundancy for Efficient Decoder-Only MLLMs","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412944615","doi":"https://doi.org/10.18653/v1/2025.findings-acl.1233"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.findings-acl.1233","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.1233","pdf_url":"https://aclanthology.org/2025.findings-acl.1233.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.findings-acl.1233.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5100442422","display_name":"Hongliang Li","orcid":"https://orcid.org/0000-0002-9821-0297"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hongliang Li","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5007567182","display_name":"Jiaxin Zhang","orcid":"https://orcid.org/0009-0008-2582-8222"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jiaxin Zhang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101829784","display_name":"Wenhui Liao","orcid":"https://orcid.org/0000-0001-8015-6918"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wenhui Liao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5012042767","display_name":"Dezhi Peng","orcid":"https://orcid.org/0000-0002-3263-3449"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dezhi Peng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5006104793","display_name":"Kai Ding","orcid":"https://orcid.org/0000-0002-2475-8639"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kai Ding","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5080674767","display_name":"Lianwen Jin","orcid":"https://orcid.org/0000-0002-5456-0957"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lianwen Jin","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":true,"cited_by_count":1,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"24056","last_page":"24067"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10388","display_name":"Advanced Steganography and Watermarking Techniques","score":0.9990000128746033,"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/T10388","display_name":"Advanced Steganography and Watermarking Techniques","score":0.9990000128746033,"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/T10901","display_name":"Advanced Data Compression Techniques","score":0.9872000217437744,"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/T10741","display_name":"Video Coding and Compression Technologies","score":0.9724000096321106,"subfield":{"id":"https://openalex.org/subfields/1711","display_name":"Signal Processing"},"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/computer-science","display_name":"Computer science","score":0.8094009160995483},{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.6703922748565674},{"id":"https://openalex.org/keywords/redundancy","display_name":"Redundancy (engineering)","score":0.6681302189826965},{"id":"https://openalex.org/keywords/computer-network","display_name":"Computer network","score":0.24771523475646973},{"id":"https://openalex.org/keywords/operating-system","display_name":"Operating system","score":0.103095144033432}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8094009160995483},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.6703922748565674},{"id":"https://openalex.org/C152124472","wikidata":"https://www.wikidata.org/wiki/Q1204361","display_name":"Redundancy (engineering)","level":2,"score":0.6681302189826965},{"id":"https://openalex.org/C31258907","wikidata":"https://www.wikidata.org/wiki/Q1301371","display_name":"Computer network","level":1,"score":0.24771523475646973},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.103095144033432}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-acl.1233","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.1233","pdf_url":"https://aclanthology.org/2025.findings-acl.1233.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.findings-acl.1233","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.1233","pdf_url":"https://aclanthology.org/2025.findings-acl.1233.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G5270964542","display_name":null,"funder_award_id":"62441604","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"},{"id":"https://openalex.org/G6561198578","display_name":null,"funder_award_id":"62476093","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":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412944615.pdf","grobid_xml":"https://content.openalex.org/works/W4412944615.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W4388335561","https://openalex.org/W2970530566","https://openalex.org/W4288261899","https://openalex.org/W4307309205","https://openalex.org/W2967478618","https://openalex.org/W4385009901","https://openalex.org/W4385572700"],"abstract_inverted_index":{"Current":[0],"Multimodal":[1],"Large":[2],"Language":[3],"Model":[4],"(MLLM)":[5],"architectures":[6,16,25,41,52],"face":[7],"a":[8,67,94,129],"critical":[9],"tradeoff":[10],"between":[11],"performance":[12,19,136],"and":[13,44,79,108],"efficiency:":[14],"decoderonly":[15],"achieve":[17,135],"higher":[18],"but":[20,29],"lower":[21,30],"efficiency,":[22],"while":[23,50,144],"cross-attentionbased":[24],"offer":[26],"greater":[27],"efficiency":[28],"performance.The":[31],"key":[32],"distinction":[33],"lies":[34],"in":[35,60,86],"how":[36],"visual":[37,48,89],"tokens":[38],"are":[39],"processed.Decoder-only":[40],"apply":[42],"self-attention":[43],"FFN":[45,78],"operations":[46],"on":[47],"tokens,":[49,90],"cross-attention":[51],"skip":[53],"these":[54,102],"computations.To":[55],"investigate":[56],"whether":[57],"redundancy":[58,110],"exists":[59],"this":[61],"computationally":[62],"expensive":[63],"process,":[64],"we":[65,134],"propose":[66],"training-free":[68,130],"framework":[69,127],"for":[70,88,101,118],"analyzing":[71],"trained":[72],"MLLMs.It":[73],"consists":[74],"of":[75],"Probe-Activated":[76],"Dynamic":[77],"Hollow":[80],"Attention,":[81],"which":[82],"enable":[83],"adjustable":[84],"reductions":[85],"computations":[87],"as":[91,93,128],"well":[92],"Layer":[95],"Ranking":[96],"Algorithm":[97],"that":[98],"prioritizes":[99],"layers":[100],"reductions.Extensive":[103],"experiments":[104],"demonstrate":[105],"substantial,":[106],"structured,":[107],"clustered":[109],"unique":[111],"to":[112,138],"decoder-only":[113],"MLLMs,":[114],"offering":[115],"valuable":[116],"insights":[117],"future":[119],"MLLM":[120],"architecture":[121],"design.Furthermore,":[122],"by":[123],"leveraging":[124],"our":[125],"reduction":[126],"inference":[131],"acceleration":[132],"approach,":[133],"comparable":[137],"or":[139],"better":[140],"than":[141],"stateof-the-art":[142],"methods":[143],"remaining":[145],"compatible":[146],"with":[147],"them.":[148]},"counts_by_year":[{"year":2025,"cited_by_count":1}],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-10-10T00:00:00"}