{"id":"https://openalex.org/W7137866501","doi":"https://doi.org/10.48550/arxiv.2603.14549","title":"ASAP: Attention-Shift-Aware Pruning for Efficient LVLM Inference","display_name":"ASAP: Attention-Shift-Aware Pruning for Efficient LVLM Inference","publication_year":2026,"publication_date":"2026-03-15","ids":{"openalex":"https://openalex.org/W7137866501","doi":"https://doi.org/10.48550/arxiv.2603.14549"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2603.14549","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.14549","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.2603.14549","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5129702209","display_name":"Surendra Pathak","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pathak, Surendra","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5129642879","display_name":"Bo Han","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Han, Bo","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":0.7139999866485596,"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":0.7139999866485596,"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":0.14169999957084656,"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":0.06279999762773514,"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/redundancy","display_name":"Redundancy (engineering)","score":0.6607000231742859},{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.6402999758720398},{"id":"https://openalex.org/keywords/pruning","display_name":"Pruning","score":0.5723000168800354},{"id":"https://openalex.org/keywords/exploit","display_name":"Exploit","score":0.5630000233650208},{"id":"https://openalex.org/keywords/set","display_name":"Set (abstract data type)","score":0.5009999871253967},{"id":"https://openalex.org/keywords/lossless-compression","display_name":"Lossless compression","score":0.4812999963760376},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.4275999963283539},{"id":"https://openalex.org/keywords/reduction","display_name":"Reduction (mathematics)","score":0.4034999907016754}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7993000149726868},{"id":"https://openalex.org/C152124472","wikidata":"https://www.wikidata.org/wiki/Q1204361","display_name":"Redundancy (engineering)","level":2,"score":0.6607000231742859},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.6402999758720398},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.5723000168800354},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.5630000233650208},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5055000185966492},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.5009999871253967},{"id":"https://openalex.org/C81081738","wikidata":"https://www.wikidata.org/wiki/Q55542","display_name":"Lossless compression","level":3,"score":0.4812999963760376},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.4275999963283539},{"id":"https://openalex.org/C111335779","wikidata":"https://www.wikidata.org/wiki/Q3454686","display_name":"Reduction (mathematics)","level":2,"score":0.4034999907016754},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.3626999855041504},{"id":"https://openalex.org/C179799912","wikidata":"https://www.wikidata.org/wiki/Q205084","display_name":"Computational complexity theory","level":2,"score":0.36079999804496765},{"id":"https://openalex.org/C129844170","wikidata":"https://www.wikidata.org/wiki/Q41299","display_name":"Quadratic equation","level":2,"score":0.33489999175071716},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.30730000138282776},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.28870001435279846},{"id":"https://openalex.org/C36503486","wikidata":"https://www.wikidata.org/wiki/Q11235244","display_name":"Domain (mathematical analysis)","level":2,"score":0.2863999903202057},{"id":"https://openalex.org/C78548338","wikidata":"https://www.wikidata.org/wiki/Q2493","display_name":"Data compression","level":2,"score":0.28220000863075256},{"id":"https://openalex.org/C36464697","wikidata":"https://www.wikidata.org/wiki/Q451553","display_name":"Visualization","level":2,"score":0.2702000141143799},{"id":"https://openalex.org/C159423971","wikidata":"https://www.wikidata.org/wiki/Q177251","display_name":"Associative property","level":2,"score":0.2694999873638153},{"id":"https://openalex.org/C164226766","wikidata":"https://www.wikidata.org/wiki/Q7293202","display_name":"Rank (graph theory)","level":2,"score":0.26820001006126404},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.2669999897480011},{"id":"https://openalex.org/C127532173","wikidata":"https://www.wikidata.org/wiki/Q179904","display_name":"Hourglass","level":2,"score":0.25699999928474426},{"id":"https://openalex.org/C2780801425","wikidata":"https://www.wikidata.org/wiki/Q5164392","display_name":"Construct (python library)","level":2,"score":0.2533999979496002},{"id":"https://openalex.org/C3826847","wikidata":"https://www.wikidata.org/wiki/Q188768","display_name":"FLOPS","level":2,"score":0.25209999084472656}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2603.14549","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.14549","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.2603.14549","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.14549","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":{"While":[0],"Large":[1],"Vision-Language":[2],"Models":[3],"(LVLMs)":[4],"demonstrate":[5],"exceptional":[6],"multi-modal":[7],"capabilities,":[8],"the":[9,47,79,91,110,130,149],"quadratic":[10],"computational":[11,156],"cost":[12],"of":[13,93,143,148],"processing":[14],"high-resolution":[15],"visual":[16,133,144],"tokens":[17,96],"remains":[18],"a":[19,65,84,118],"critical":[20],"bottleneck.":[21],"Though":[22],"recent":[23],"token":[24,41,56,111],"reduction":[25],"strategies":[26],"attempt":[27],"to":[28,39],"accelerate":[29],"inference,":[30],"such":[31],"methods":[32],"inadequately":[33],"exploit":[34],"attention":[35,57,80,88],"values":[36],"and":[37],"fail":[38],"address":[40],"redundancy.":[42],"More":[43],"critically,":[44],"they":[45],"overlook":[46],"``attention":[48],"shift''":[49],"phenomenon":[50],"inherent":[51],"in":[52],"LVLMs,":[53],"which":[54],"skews":[55],"scores.":[58],"In":[59],"this":[60],"work,":[61],"we":[62,77,103],"propose":[63],"ASAP,":[64],"novel":[66],"training-free,":[67],"KV-Cache-compatible":[68],"pruning":[69],"recipe":[70],"that":[71,105,123],"comprehensively":[72],"addresses":[73],"these":[74],"limitations.":[75],"First,":[76],"mitigate":[78],"shift":[81],"by":[82,158],"utilizing":[83],"dynamic":[85],"bidirectional":[86],"soft":[87,120],"mask,":[89],"ensuring":[90],"selection":[92],"genuinely":[94],"informative":[95],"rather":[97],"than":[98],"naive":[99],"attention-based":[100],"selection.":[101],"Second,":[102],"posit":[104],"high":[106],"semantic":[107],"redundancy":[108],"within":[109],"set":[112],"degrades":[113],"performance.":[114],"We":[115],"therefore":[116],"introduce":[117],"weighted":[119],"merging":[121],"component":[122],"merges":[124],"semantically":[125],"similar":[126],"tokens,":[127],"preserving":[128],"only":[129],"most":[131],"feature-dense":[132],"patches":[134],"for":[135],"subsequent":[136],"layers.":[137],"ASAP":[138],"achieves":[139],"virtually":[140],"lossless":[141],"compression":[142],"context,":[145],"retaining":[146],"99.02%":[147],"original":[150],"LLaVA-NeXT-7B":[151],"performance":[152],"while":[153],"aggressively":[154],"slashing":[155],"FLOPs":[157],"~80%.":[159]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-03-18T00:00:00"}