{"id":"https://openalex.org/W7162088579","doi":"https://doi.org/10.48550/arxiv.2605.21642","title":"Ablate-to-Validate: Are Vision-Language Models Really Using Continuous Thought Tokens?","display_name":"Ablate-to-Validate: Are Vision-Language Models Really Using Continuous Thought Tokens?","publication_year":2026,"publication_date":"2026-05-20","ids":{"openalex":"https://openalex.org/W7162088579","doi":"https://doi.org/10.48550/arxiv.2605.21642"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.21642","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.21642","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"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":null,"license_id":null,"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.2605.21642","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5136759873","display_name":"Tianyi Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Tianyi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5030320952","display_name":"Mahtab Bigverdi","orcid":"https://orcid.org/0000-0001-9336-7570"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bigverdi, Mahtab","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5136774115","display_name":"Ranjay Krishna","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Krishna, Ranjay","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":0.954800009727478,"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":0.954800009727478,"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":0.006300000008195639,"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":0.004800000227987766,"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/security-token","display_name":"Security token","score":0.9284999966621399},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.5509999990463257},{"id":"https://openalex.org/keywords/oracle","display_name":"Oracle","score":0.5314000248908997},{"id":"https://openalex.org/keywords/task","display_name":"Task (project management)","score":0.4634000062942505},{"id":"https://openalex.org/keywords/decoding-methods","display_name":"Decoding methods","score":0.44760000705718994},{"id":"https://openalex.org/keywords/proxy","display_name":"Proxy (statistics)","score":0.35249999165534973}],"concepts":[{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.9284999966621399},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.685699999332428},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.5509999990463257},{"id":"https://openalex.org/C55166926","wikidata":"https://www.wikidata.org/wiki/Q2892946","display_name":"Oracle","level":2,"score":0.5314000248908997},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.4634000062942505},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.44760000705718994},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4465999901294708},{"id":"https://openalex.org/C2780148112","wikidata":"https://www.wikidata.org/wiki/Q1432581","display_name":"Proxy (statistics)","level":2,"score":0.35249999165534973},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.3402999937534332},{"id":"https://openalex.org/C127162648","wikidata":"https://www.wikidata.org/wiki/Q16858953","display_name":"Channel (broadcasting)","level":2,"score":0.3393000066280365},{"id":"https://openalex.org/C118505674","wikidata":"https://www.wikidata.org/wiki/Q42586063","display_name":"Encoder","level":2,"score":0.31130000948905945},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.30630001425743103},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2985000014305115},{"id":"https://openalex.org/C115067241","wikidata":"https://www.wikidata.org/wiki/Q1639854","display_name":"Token passing","level":3,"score":0.296099990606308},{"id":"https://openalex.org/C2780385302","wikidata":"https://www.wikidata.org/wiki/Q367158","display_name":"Protocol (science)","level":3,"score":0.2928999960422516},{"id":"https://openalex.org/C155076860","wikidata":"https://www.wikidata.org/wiki/Q1932004","display_name":"Token economy","level":3,"score":0.28630000352859497},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.2646999955177307}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.21642","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.21642","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.21642","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.21642","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"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":null,"license_id":null,"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":{"Vision-language":[0],"models":[1,27,126],"(VLMs)":[2],"are":[3,168],"increasingly":[4],"augmented":[5],"with":[6,93,121],"continuous":[7,131,216],"or":[8,47,97,107,132,185],"latent":[9,194],"non-textual":[10],"tokens":[11,31,92],"intended":[12],"to":[13,127,150],"support":[14],"\"visual":[15],"thinking.\"":[16],"Despite":[17],"improved":[18],"task":[19],"accuracy,":[20],"this":[21],"alone":[22],"does":[23],"not":[24],"show":[25],"that":[26],"actually":[28],"use":[29],"these":[30],"for":[32,56,172,212],"reasoning":[33,120],"--":[34],"gains":[35,167],"may":[36],"arise":[37],"from":[38],"confounds":[39],"such":[40],"as":[41,67,199,206],"added":[42],"context":[43],"length,":[44],"special-token":[45],"anchoring,":[46],"training-time":[48],"regularization.":[49],"We":[50,146,203],"formalize":[51],"a":[52,73,113,169,188,193,207],"diagnostic":[53,209],"principle,":[54],"Ablate-to-Validate,":[55],"testing":[57],"whether":[58,101],"latent-token":[59,173],"content":[60,106,182],"is":[61,183],"genuinely":[62],"utilized,":[63],"and":[64,86,123,129,143,161,196],"instantiate":[65],"it":[66,198],"the":[68,81],"Token":[69],"Replacement":[70],"Test":[71],"(TRT),":[72],"standardized":[74],"suite":[75],"of":[76],"content-replacement":[77],"ablations.":[78],"TRT":[79,149,205],"holds":[80],"prompt,":[82],"image,":[83],"token":[84,105,110,144,181],"budget,":[85],"decoding":[87],"fixed":[88],"while":[89],"replacing":[90],"intermediate":[91],"zero,":[94],"random,":[95],"first-repeat,":[96],"oracle":[98],"alternatives,":[99],"isolating":[100],"performance":[102],"depends":[103],"on":[104,109,158],"merely":[108],"presence.":[111],"As":[112],"controlled":[114],"testbed,":[115],"we":[116],"study":[117],"relative":[118],"depth":[119,134],"LLaVA-13B":[122],"Qwen2.5-VL-3B,":[124],"training":[125],"predict":[128],"consume":[130],"discrete":[133],"spans":[135],"across":[136],"multiple":[137],"frozen":[138],"encoders":[139],"(SigLIP2,":[140],"CLIP,":[141],"DINOv2)":[142],"budgets.":[145],"additionally":[147],"apply":[148],"three":[151],"off-the-shelf":[152],"visual-thinking":[153],"systems":[154],"(Mirage,":[155],"Mull-Tokens,":[156],"CoVT)":[157],"BLINK,":[159],"VSP,":[160],"CV-Bench.":[162],"Across":[163],"all":[164],"settings,":[165],"accuracy":[166,211],"misleading":[170],"proxy":[171],"reasoning:":[174],"VLMs":[175],"retain":[176],"most":[177],"improvement":[178],"even":[179],"when":[180],"corrupted":[184],"replaced,":[186],"revealing":[187],"persistent":[189],"gap":[190],"between":[191],"having":[192],"channel":[195],"using":[197],"an":[200],"information":[201],"bottleneck.":[202],"recommend":[204],"standard":[208],"alongside":[210],"any":[213],"method":[214],"introducing":[215],"thought":[217],"tokens.":[218]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-23T00:00:00"}