{"id":"https://openalex.org/W7202229253","doi":"https://doi.org/10.48550/arxiv.2608.09176","title":"Not All Visual Tokens Are Equally Safe to Remove:Consequence-Sensitive Visual Token Compression","display_name":"Not All Visual Tokens Are Equally Safe to Remove:Consequence-Sensitive Visual Token Compression","publication_year":2026,"publication_date":"2026-08-10","ids":{"openalex":"https://openalex.org/W7202229253","doi":"https://doi.org/10.48550/arxiv.2608.09176"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2608.09176","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.09176","pdf_url":"https://arxiv.org/pdf/2608.09176","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2608.09176","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5137653648","display_name":"Jingbo Wen","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I129604602","display_name":"The University of Sydney","ror":"https://ror.org/0384j8v12","country_code":"AU","type":"education","lineage":["https://openalex.org/I129604602"]}],"countries":["AU"],"is_corresponding":false,"raw_author_name":"Wen, Jingbo","raw_affiliation_strings":["The University of Sydney"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The University of Sydney","institution_ids":["https://openalex.org/I129604602"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137981095","display_name":"Liang He","orcid":"https://orcid.org/0000-0002-4723-5486","observed_orcids":["https://orcid.org/0000-0002-4723-5486"]},"institutions":[{"id":"https://openalex.org/I116953780","display_name":"Tongji University","ror":"https://ror.org/03rc6as71","country_code":"CN","type":"education","lineage":["https://openalex.org/I116953780"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"He, Liang","raw_affiliation_strings":["Tongji University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Tongji University","institution_ids":["https://openalex.org/I116953780"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5043290825","display_name":"Mingyu Cao","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I28290843","display_name":"University of Surrey","ror":"https://ror.org/00ks66431","country_code":"GB","type":"education","lineage":["https://openalex.org/I28290843"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Cao, Mingyu","raw_affiliation_strings":["University of Surrey"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Surrey","institution_ids":["https://openalex.org/I28290843"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5146492119","display_name":"Haoyu Wang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I205237279","display_name":"Nankai University","ror":"https://ror.org/01y1kjr75","country_code":"CN","type":"education","lineage":["https://openalex.org/I205237279"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Wang, Haoyu","raw_affiliation_strings":["Nankai University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Nankai University","institution_ids":["https://openalex.org/I205237279"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134775557","display_name":"Minxuan Hu","orcid":"https://orcid.org/0009-0007-1733-0099","observed_orcids":["https://orcid.org/0009-0007-1733-0099"]},"institutions":[{"id":"https://openalex.org/I205783295","display_name":"Cornell University","ror":"https://ror.org/05bnh6r87","country_code":"US","type":"education","lineage":["https://openalex.org/I205783295"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Hu, Minxuan","raw_affiliation_strings":["Cornell University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Cornell University","institution_ids":["https://openalex.org/I205783295"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5146529162","display_name":"Kangning Cui","orcid":"https://orcid.org/0000-0002-1856-5064","observed_orcids":["https://orcid.org/0000-0002-1856-5064"]},"institutions":[{"id":"https://openalex.org/I168719708","display_name":"City University of Hong Kong","ror":"https://ror.org/03q8dnn23","country_code":"HK","type":"education","lineage":["https://openalex.org/I168719708"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Cui, Kangning","raw_affiliation_strings":["City University of Hong Kong"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"City University of Hong Kong","institution_ids":["https://openalex.org/I168719708"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5146517024","display_name":"Xilu Wang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I28290843","display_name":"University of Surrey","ror":"https://ror.org/00ks66431","country_code":"GB","type":"education","lineage":["https://openalex.org/I28290843"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Wang, Xilu","raw_affiliation_strings":["University of Surrey"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Surrey","institution_ids":["https://openalex.org/I28290843"]}]}],"institutions":[],"countries_distinct_count":5,"institutions_distinct_count":6,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":3,"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":0.951200008392334,"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":0.951200008392334,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.039400000125169754,"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/T10601","display_name":"Handwritten Text Recognition Techniques","score":0.0017000000225380063,"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/visual-token-compression","display_name":"visual token compression","score":0.9819999933242798},{"id":"https://openalex.org/keywords/vision-language-models","display_name":"vision-language models","score":0.859000027179718},{"id":"https://openalex.org/keywords/error-costs","display_name":"error costs","score":0.4399999976158142},{"id":"https://openalex.org/keywords/calibration","display_name":"calibration","score":0.36000001430511475},{"id":"https://openalex.org/keywords/token-selection","display_name":"token selection","score":0.5389999747276306},{"id":"https://openalex.org/keywords/inference-latency","display_name":"inference latency","score":0.5109999775886536},{"id":"https://openalex.org/keywords/document-image-understanding","display_name":"document image understanding","score":0.3449999988079071}],"concepts":[{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.8442000150680542},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7506999969482422},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.48649999499320984},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.40709999203681946},{"id":"https://openalex.org/C162262903","wikidata":"https://www.wikidata.org/wiki/Q343527","display_name":"Allocator","level":2,"score":0.4041000008583069},{"id":"https://openalex.org/C2781089630","wikidata":"https://www.wikidata.org/wiki/Q21856745","display_name":"Realization (probability)","level":2,"score":0.3716999888420105},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.36800000071525574},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.35659998655319214},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.3257000148296356},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3125},{"id":"https://openalex.org/C28719098","wikidata":"https://www.wikidata.org/wiki/Q44946","display_name":"Point (geometry)","level":2,"score":0.3116999864578247},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.30489999055862427},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.2957000136375427},{"id":"https://openalex.org/C2986053828","wikidata":"https://www.wikidata.org/wiki/Q355217","display_name":"Time budget","level":2,"score":0.29269999265670776},{"id":"https://openalex.org/C2776175482","wikidata":"https://www.wikidata.org/wiki/Q1195816","display_name":"Transfer (computing)","level":2,"score":0.28929999470710754},{"id":"https://openalex.org/C2780385302","wikidata":"https://www.wikidata.org/wiki/Q367158","display_name":"Protocol (science)","level":3,"score":0.27410000562667847},{"id":"https://openalex.org/C36464697","wikidata":"https://www.wikidata.org/wiki/Q451553","display_name":"Visualization","level":2,"score":0.2590999901294708},{"id":"https://openalex.org/C2780233690","wikidata":"https://www.wikidata.org/wiki/Q535347","display_name":"Transparency (behavior)","level":2,"score":0.2572000026702881},{"id":"https://openalex.org/C115067241","wikidata":"https://www.wikidata.org/wiki/Q1639854","display_name":"Token passing","level":3,"score":0.2531000077724457},{"id":"https://openalex.org/C78548338","wikidata":"https://www.wikidata.org/wiki/Q2493","display_name":"Data compression","level":2,"score":0.2515000104904175}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2608.09176","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.09176","pdf_url":"https://arxiv.org/pdf/2608.09176","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2608.09176","doi":"https://doi.org/10.48550/arxiv.2608.09176","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2608.09176","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,"listed_in":[],"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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2608.09176","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.09176","pdf_url":"https://arxiv.org/pdf/2608.09176","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7202229253.pdf","grobid_xml":"https://content.openalex.org/works/W7202229253.grobid-xml"},"referenced_works_count":10,"referenced_works":["https://openalex.org/W4391556981","https://openalex.org/W4403370767","https://openalex.org/W4407794264","https://openalex.org/W4415037622","https://openalex.org/W4417032857","https://openalex.org/W7101897019","https://openalex.org/W7134924069","https://openalex.org/W7160959899","https://openalex.org/W7163566988","https://openalex.org/W7164343917"],"related_works":["https://openalex.org/W4404356453","https://openalex.org/W7162751813","https://openalex.org/W7138182710","https://openalex.org/W7220991220","https://openalex.org/W7131400887","https://openalex.org/W7204827344","https://openalex.org/W7162759614","https://openalex.org/W4417288840","https://openalex.org/W7161585240","https://openalex.org/W7163191770"],"abstract_inverted_index":{"Visual":[0],"token":[1,69,98,154,172,193,230],"compression":[2],"for":[3],"vision--language":[4],"models":[5],"(VLMs)":[6],"has":[7],"largely":[8],"relied":[9],"on":[10,42],"criteria":[11],"such":[12],"as":[13,201],"attention,":[14],"redundancy,":[15],"and":[16,94,116,192,222,228],"uncertainty":[17],"to":[18,78,135,148],"maximize":[19],"average":[20],"accuracy":[21],"under":[22,150],"a":[23,59,86,111,157,234],"fixed":[24],"compute":[25],"budget,":[26,155],"implicitly":[27],"assuming":[28],"that":[29],"all":[30],"errors":[31,145,188],"carry":[32],"equal":[33],"cost.":[34],"However,":[35],"the":[36,96,122,151,202],"consequence":[37,102],"of":[38],"an":[39,49,180],"incorrect":[40],"prediction":[41],"downstream":[43],"tasks":[44],"is":[45,185],"rarely":[46],"symmetric:":[47],"misreading":[48],"invoice":[50],"amount":[51],"can":[52],"be":[53],"far":[54],"more":[55],"costly":[56,134],"than":[57,163,251],"misclassifying":[58],"background":[60],"color.":[61],"Motivated":[62],"by":[63,243],"this,":[64],"we":[65,178],"introduce":[66],"consequence-sensitive":[67,238],"visual":[68,73],"compression,":[70],"which":[71,131],"allocates":[72],"computation":[74],"across":[75,174,211],"requests":[76],"according":[77],"their":[79],"potential":[80],"error":[81,168,242],"costs.":[82],"Our":[83],"method":[84,142],"follows":[85],"calibrate-then-allocate":[87],"procedure,":[88],"estimating":[89],"consequence-specific":[90],"error-budget":[91],"curves":[92],"offline":[93],"applying":[95],"calibrated":[97],"budgets":[99],"online":[100],"using":[101],"signals":[103],"available":[104],"from":[105,121,146],"question":[106],"or":[107],"task":[108],"information.":[109],"On":[110,233],"controlled":[112],"within-task":[113],"benchmark,":[114],"high-":[115],"low-consequence":[117],"questions":[118,132,197],"are":[119,133,189],"drawn":[120],"same":[123,152],"document":[124],"images,":[125],"so":[126],"content":[127],"alone":[128],"cannot":[129],"reveal":[130],"get":[136],"wrong.":[137],"In":[138],"this":[139],"setting,":[140],"our":[141],"reduces":[143,240],"high-stakes":[144],"0.300":[147],"0.133":[149],"total":[153],"whereas":[156],"content-driven":[158],"allocator":[159],"performs":[160],"no":[161],"better":[162],"uniform":[164,183],"allocation.":[165],"Measuring":[166],"how":[167],"rates":[169],"change":[170],"with":[171],"budget":[173,217],"different":[175],"cost":[176,203],"ratios,":[177],"derive":[179],"allocation":[181,184,207,239],"frontier:":[182],"optimal":[186],"when":[187],"equally":[190],"costly,":[191],"transfer":[194],"toward":[195],"high-consequence":[196],"becomes":[198],"increasingly":[199],"beneficial":[200],"gap":[204],"grows.":[205],"This":[206],"principle":[208],"generalizes":[209],"well":[210],"three":[212],"dense":[213],"vision-language":[214],"benchmarks,":[215],"two":[216,225],"realization":[218],"mechanisms":[219],"(token":[220],"deletion":[221],"resolution":[223],"reallocation),":[224],"VLM":[226],"architectures,":[227],"multiple":[229],"selection":[231],"strategies.":[232],"realistic":[235],"mixed":[236],"workload,":[237],"cost-weighted":[241],"38%":[244],"while":[245],"achieving":[246],"approximately":[247],"21%":[248],"lower":[249],"latency":[250],"full-resolution":[252],"inference.":[253]},"counts_by_year":[{"year":2026,"cited_by_count":3}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-08-12T00:00:00"}