{"id":"https://openalex.org/W7164020727","doi":"https://doi.org/10.48550/arxiv.2606.08511","title":"Look Less, Reason More: Block-wise Attention Skipping for Efficient Multimodal LLMs","display_name":"Look Less, Reason More: Block-wise Attention Skipping for Efficient Multimodal LLMs","publication_year":2026,"publication_date":"2026-06-07","ids":{"openalex":"https://openalex.org/W7164020727","doi":"https://doi.org/10.48550/arxiv.2606.08511"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.08511","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.08511","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","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.2606.08511","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5138231388","display_name":"Jie Ma","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ma, Jie","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5117839877","display_name":"Zhike Qiu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qiu, Zhike","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138221245","display_name":"Jiayi Ji","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ji, Jiayi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138229959","display_name":"Xiaoshuai Sun","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sun, Xiaoshuai","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5138260946","display_name":"Rongrong Ji","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ji, Rongrong","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":0.954200029373169,"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":0.954200029373169,"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":0.010700000450015068,"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.007300000172108412,"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":0.6958000063896179},{"id":"https://openalex.org/keywords/bottleneck","display_name":"Bottleneck","score":0.6651999950408936},{"id":"https://openalex.org/keywords/rendering","display_name":"Rendering (computer graphics)","score":0.5900999903678894},{"id":"https://openalex.org/keywords/visual-reasoning","display_name":"Visual reasoning","score":0.49720001220703125},{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.44859999418258667},{"id":"https://openalex.org/keywords/inefficiency","display_name":"Inefficiency","score":0.421999990940094},{"id":"https://openalex.org/keywords/visualization","display_name":"Visualization","score":0.3718000054359436},{"id":"https://openalex.org/keywords/semantics","display_name":"Semantics (computer science)","score":0.3474000096321106}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7832000255584717},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6958000063896179},{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.6651999950408936},{"id":"https://openalex.org/C205711294","wikidata":"https://www.wikidata.org/wiki/Q176953","display_name":"Rendering (computer graphics)","level":2,"score":0.5900999903678894},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5540000200271606},{"id":"https://openalex.org/C2777508537","wikidata":"https://www.wikidata.org/wiki/Q7936620","display_name":"Visual reasoning","level":2,"score":0.49720001220703125},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.44859999418258667},{"id":"https://openalex.org/C2778869765","wikidata":"https://www.wikidata.org/wiki/Q6028363","display_name":"Inefficiency","level":2,"score":0.421999990940094},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3756999969482422},{"id":"https://openalex.org/C36464697","wikidata":"https://www.wikidata.org/wiki/Q451553","display_name":"Visualization","level":2,"score":0.3718000054359436},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.3474000096321106},{"id":"https://openalex.org/C107457646","wikidata":"https://www.wikidata.org/wiki/Q207434","display_name":"Human\u2013computer interaction","level":1,"score":0.31610000133514404},{"id":"https://openalex.org/C44291984","wikidata":"https://www.wikidata.org/wiki/Q1074173","display_name":"Question answering","level":2,"score":0.30799999833106995},{"id":"https://openalex.org/C2986089797","wikidata":"https://www.wikidata.org/wiki/Q6501338","display_name":"Visual attention","level":3,"score":0.2953000068664551},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.2858999967575073},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.27619999647140503},{"id":"https://openalex.org/C174348530","wikidata":"https://www.wikidata.org/wiki/Q188635","display_name":"Bridging (networking)","level":2,"score":0.2736000120639801},{"id":"https://openalex.org/C2780598303","wikidata":"https://www.wikidata.org/wiki/Q65921492","display_name":"Flexibility (engineering)","level":2,"score":0.2709999978542328},{"id":"https://openalex.org/C206588197","wikidata":"https://www.wikidata.org/wiki/Q846574","display_name":"Reuse","level":2,"score":0.27090001106262207},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.2676999866962433},{"id":"https://openalex.org/C66024118","wikidata":"https://www.wikidata.org/wiki/Q1122506","display_name":"Computational model","level":2,"score":0.26589998602867126}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.08511","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.08511","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2606.08511","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.08511","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"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":{"Multimodal":[0],"Large":[1],"Language":[2],"Models":[3],"(MLLMs)":[4],"face":[5],"a":[6,26,86,125,151],"significant":[7],"inference":[8,88],"bottleneck":[9],"due":[10],"to":[11,129,146,153,164,172,181],"the":[12,74,132,185],"quadratic":[13],"computational":[14],"cost":[15],"of":[16],"self-attention":[17,54,111],"over":[18],"long":[19],"visual":[20,39,71,110],"token":[21],"sequences.":[22],"However,":[23],"we":[24,82,161],"identify":[25],"critical":[27],"inefficiency":[28],"in":[29,49,55,64],"current":[30],"architectures:":[31],"Visual":[32],"Attention":[33],"Saturation.":[34],"Our":[35],"analysis":[36],"reveals":[37],"that":[38,90,115,139,163],"tokens":[40],"rapidly":[41],"establish":[42],"their":[43],"spatial":[44,92],"structure":[45],"and":[46],"intra-modal":[47],"relationships":[48],"early":[50],"layers,":[51],"rendering":[52],"visual-to-visual":[53],"deeper":[56],"layers":[57,66],"computationally":[58],"redundant.":[59],"Conversely,":[60],"Feed-Forward":[61],"Networks":[62],"(FFNs)":[63],"these":[65],"remain":[67],"essential":[68],"for":[69],"projecting":[70],"features":[72],"into":[73],"evolving":[75],"textual":[76],"semantic":[77,95],"space.":[78],"Leveraging":[79],"this":[80],"insight,":[81],"present":[83],"Visual-Skip":[84],"(V-Skip),":[85],"training-free":[87],"paradigm":[89],"decouples":[91],"interaction":[93],"from":[94],"evolution.":[96],"Rather":[97],"than":[98],"discarding":[99],"tokens,":[100],"V-Skip":[101,123,140],"imposes":[102],"block-wise":[103,148],"structured":[104],"sparsity":[105,134],"by":[106],"selectively":[107],"bypassing":[108],"saturated":[109],"modules.":[112],"Furthermore,":[113],"recognizing":[114],"varying":[116],"downstream":[117],"tasks":[118],"demand":[119],"distinct":[120],"reasoning":[121],"depths,":[122],"employs":[124],"lightweight,":[126],"few-shot":[127],"calibration":[128],"dynamically":[130],"route":[131],"task-optimal":[133],"path.":[135],"Extensive":[136],"experiments":[137],"demonstrate":[138],"effectively":[141],"bypasses":[142],"redundant":[143],"vision":[144],"attention":[145],"achieve":[147],"sparsity,":[149],"maintaining":[150],"94.16%":[152],"100.31%":[154],"performance":[155],"retention":[156],"across":[157],"diverse":[158],"MLLMs.":[159],"Ultimately,":[160],"prove":[162],"reason":[165],"more":[166],"effectively,":[167],"models":[168],"do":[169],"not":[170],"need":[171,180],"discard":[173],"what":[174],"they":[175,178],"see":[176],"--":[177],"simply":[179],"\"look":[182],"less\"":[183],"at":[184],"right":[186],"depth.":[187]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-06-10T00:00:00"}