{"id":"https://openalex.org/W4416640809","doi":"https://doi.org/10.1016/j.neucom.2026.134676","title":"Interpreting in-context learning in vision-language models for semantics\u2013statistics disentanglement via out-of-distribution benchmark","display_name":"Interpreting in-context learning in vision-language models for semantics\u2013statistics disentanglement via out-of-distribution benchmark","publication_year":2026,"publication_date":"2026-08-03","ids":{"openalex":"https://openalex.org/W4416640809","doi":"https://doi.org/10.1016/j.neucom.2026.134676"},"language":"en","primary_location":{"id":"doi:10.1016/j.neucom.2026.134676","doi":"https://doi.org/10.1016/j.neucom.2026.134676","is_oa":true,"landing_page_url":"https://doi.org/10.1016/j.neucom.2026.134676","pdf_url":null,"source":{"id":"https://openalex.org/S45693802","display_name":"Neurocomputing","issn_l":"0925-2312","issn":["0925-2312","1872-8286"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["anvur-scientific","ccf-c","cwts-core","dongbi-b","doyens","jufo-2","ki-jl-2","norway-2","poland-140","russia-white-list-1","vabb-shw"],"host_organization":"https://openalex.org/P4310320990","host_organization_name":"Elsevier BV","host_organization_lineage":["https://openalex.org/P4310320990"],"host_organization_lineage_names":["Elsevier BV"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Neurocomputing","raw_type":"journal-article"},"volume":null,"type":"article","is_traditional":true,"indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"hybrid","oa_url":"https://doi.org/10.1016/j.neucom.2026.134676","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5078415335","display_name":"Yosuke Miyanishi","orcid":"https://orcid.org/0009-0002-0021-006X","observed_orcids":["https://orcid.org/0009-0002-0021-006X"]},"institutions":[{"id":"https://openalex.org/I4210089607","display_name":"CyberAgent (Japan)","ror":"https://ror.org/0060jg679","country_code":"JP","type":"company","lineage":["https://openalex.org/I4210089607"]}],"countries":["JP"],"is_corresponding":true,"raw_author_name":"Yosuke Miyanishi","raw_affiliation_strings":["CyberAgent Inc, Shibuya City, Tokyo, Japan"],"raw_orcid":"https://orcid.org/0009-0002-0021-006X","affiliations":[{"raw_affiliation_string":"CyberAgent Inc, Shibuya City, Tokyo, Japan","institution_ids":["https://openalex.org/I4210089607"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5077641909","display_name":"Le-Minh Nguyen","orcid":"https://orcid.org/0000-0002-2265-1010","observed_orcids":["https://orcid.org/0000-0002-2265-1010"]},"institutions":[{"id":"https://openalex.org/I177738480","display_name":"Japan Advanced Institute of Science and Technology","ror":"https://ror.org/03frj4r98","country_code":"JP","type":"education","lineage":["https://openalex.org/I177738480"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Nguyen Le Minh","raw_affiliation_strings":["Japan Advanced Institute of Science and Technology (JAIST), Nomi, Ishikawa Prefecture, Japan"],"raw_orcid":"https://orcid.org/0000-0002-2265-1010","affiliations":[{"raw_affiliation_string":"Japan Advanced Institute of Science and Technology (JAIST), Nomi, Ishikawa Prefecture, Japan","institution_ids":["https://openalex.org/I177738480"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":["https://openalex.org/A5078415335"],"corresponding_institution_ids":["https://openalex.org/I4210089607"],"apc_list":{"value":2670,"currency":"USD","value_usd":2670},"apc_paid":{"value":2670,"currency":"USD","value_usd":2670},"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.00581491,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":"703","issue":null,"first_page":"134676","last_page":"134676"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.9467999935150146,"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":0.9467999935150146,"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":0.05050000175833702,"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/T10028","display_name":"Topic Modeling","score":0.0008999999845400453,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/in-context-learning","display_name":"in-context learning","score":1.0},{"id":"https://openalex.org/keywords/vision-language-models","display_name":"vision-language models","score":0.996999979019165},{"id":"https://openalex.org/keywords/out-of-distribution-generalization","display_name":"out-of-distribution generalization","score":0.9449999928474426},{"id":"https://openalex.org/keywords/distribution-shift","display_name":"distribution shift","score":0.8960000276565552},{"id":"https://openalex.org/keywords/multimodal-representation-learning","display_name":"multimodal representation learning","score":0.6179999709129333},{"id":"https://openalex.org/keywords/llava","display_name":"LLaVA","score":0.6570000052452087},{"id":"https://openalex.org/keywords/visual-question-answering","display_name":"visual question answering","score":0.8320000171661377},{"id":"https://openalex.org/keywords/label-bias","display_name":"label bias","score":0.6320000290870667},{"id":"https://openalex.org/keywords/hateful-memes","display_name":"hateful memes","score":0.6840000152587891},{"id":"https://openalex.org/keywords/zero-shot-learning","display_name":"zero-shot learning","score":0.31299999356269836}],"concepts":[{"id":"https://openalex.org/C97256817","wikidata":"https://www.wikidata.org/wiki/Q1462316","display_name":"Spurious relationship","level":2,"score":0.6988000273704529},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6875},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.6201000213623047},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.6114000082015991},{"id":"https://openalex.org/C108650721","wikidata":"https://www.wikidata.org/wiki/Q1783253","display_name":"Counterfactual thinking","level":2,"score":0.5439000129699707},{"id":"https://openalex.org/C2778828372","wikidata":"https://www.wikidata.org/wiki/Q5283209","display_name":"Distributional semantics","level":3,"score":0.5428000092506409},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.4754999876022339},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.46389999985694885},{"id":"https://openalex.org/C774472","wikidata":"https://www.wikidata.org/wiki/Q6760393","display_name":"Margin (machine learning)","level":2,"score":0.43160000443458557},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.43149998784065247},{"id":"https://openalex.org/C2776359362","wikidata":"https://www.wikidata.org/wiki/Q2145286","display_name":"Representation (politics)","level":3,"score":0.4260999858379364},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4178999960422516},{"id":"https://openalex.org/C192209626","wikidata":"https://www.wikidata.org/wiki/Q190909","display_name":"Focus (optics)","level":2,"score":0.4034999907016754},{"id":"https://openalex.org/C5274069","wikidata":"https://www.wikidata.org/wiki/Q2285707","display_name":"Categorical variable","level":2,"score":0.365200012922287},{"id":"https://openalex.org/C130318100","wikidata":"https://www.wikidata.org/wiki/Q2268914","display_name":"Semantic similarity","level":2,"score":0.34630000591278076},{"id":"https://openalex.org/C149441793","wikidata":"https://www.wikidata.org/wiki/Q200726","display_name":"Probability distribution","level":2,"score":0.3441999852657318},{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.33970001339912415},{"id":"https://openalex.org/C114289077","wikidata":"https://www.wikidata.org/wiki/Q3284399","display_name":"Statistical model","level":2,"score":0.3165000081062317},{"id":"https://openalex.org/C103278499","wikidata":"https://www.wikidata.org/wiki/Q254465","display_name":"Similarity (geometry)","level":3,"score":0.3084000051021576},{"id":"https://openalex.org/C2780586882","wikidata":"https://www.wikidata.org/wiki/Q7520643","display_name":"Simple (philosophy)","level":2,"score":0.26980000734329224},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.2590000033378601},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.2563000023365021},{"id":"https://openalex.org/C2776502983","wikidata":"https://www.wikidata.org/wiki/Q690182","display_name":"Contrast (vision)","level":2,"score":0.2526000142097473}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.1016/j.neucom.2026.134676","doi":"https://doi.org/10.1016/j.neucom.2026.134676","is_oa":true,"landing_page_url":"https://doi.org/10.1016/j.neucom.2026.134676","pdf_url":null,"source":{"id":"https://openalex.org/S45693802","display_name":"Neurocomputing","issn_l":"0925-2312","issn":["0925-2312","1872-8286"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["anvur-scientific","ccf-c","cwts-core","dongbi-b","doyens","jufo-2","ki-jl-2","norway-2","poland-140","russia-white-list-1","vabb-shw"],"host_organization":"https://openalex.org/P4310320990","host_organization_name":"Elsevier BV","host_organization_lineage":["https://openalex.org/P4310320990"],"host_organization_lineage_names":["Elsevier BV"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Neurocomputing","raw_type":"journal-article"},{"id":"doi:10.2139/ssrn.5800458","doi":"https://doi.org/10.2139/ssrn.5800458","is_oa":true,"landing_page_url":"https://doi.org/10.2139/ssrn.5800458","pdf_url":null,"source":{"id":"https://openalex.org/S4210172589","display_name":"SSRN Electronic Journal","issn_l":"1556-5068","issn":["1556-5068"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I1318003438","host_organization_name":"RELX Group (Netherlands)","host_organization_lineage":["https://openalex.org/I1318003438"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"posted-content"}],"best_oa_location":{"id":"doi:10.1016/j.neucom.2026.134676","doi":"https://doi.org/10.1016/j.neucom.2026.134676","is_oa":true,"landing_page_url":"https://doi.org/10.1016/j.neucom.2026.134676","pdf_url":null,"source":{"id":"https://openalex.org/S45693802","display_name":"Neurocomputing","issn_l":"0925-2312","issn":["0925-2312","1872-8286"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["anvur-scientific","ccf-c","cwts-core","dongbi-b","doyens","jufo-2","ki-jl-2","norway-2","poland-140","russia-white-list-1","vabb-shw"],"host_organization":"https://openalex.org/P4310320990","host_organization_name":"Elsevier BV","host_organization_lineage":["https://openalex.org/P4310320990"],"host_organization_lineage_names":["Elsevier BV"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Neurocomputing","raw_type":"journal-article"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":2,"referenced_works":["https://openalex.org/W2019655958","https://openalex.org/W3134095442"],"related_works":["https://openalex.org/W4393027600","https://openalex.org/W4417136328","https://openalex.org/W4410915147","https://openalex.org/W7102414624","https://openalex.org/W7223318781","https://openalex.org/W7172536070","https://openalex.org/W7221504837","https://openalex.org/W7160300696","https://openalex.org/W4379473869","https://openalex.org/W4386794522"],"abstract_inverted_index":{"Understanding":[0],"when":[1,55,234,248],"and":[2,41,59,114,121,139,152,177,188,204,215,236,254],"why":[3],"in-context":[4],"learning":[5],"(ICL)":[6],"helps":[7,250],"vision-language":[8],"models":[9,45],"(VLMs)":[10],"is":[11,34,175,185],"central":[12],"to":[13,61],"building":[14],"robust":[15],"multimodal":[16],"systems":[17],"under":[18,97,251],"distribution":[19,252],"shift.":[20],"We":[21,100,126],"focus":[22],"on":[23,144,156,208],"out-of-distribution":[24],"(OoD)":[25],"settings":[26],"because":[27],"deployments":[28],"rarely":[29],"match":[30],"training":[31],"data,":[32],"ICL":[33,56,80,237,249],"often":[35],"the":[36,171,228],"first":[37],"line":[38],"of":[39,247],"adaptation,":[40],"OoD":[42,98,146],"stress-tests":[43],"whether":[44],"use":[46],"semantics":[47,64,94],"or":[48,78,180],"spurious":[49],"format":[50],"cues.":[51],"This":[52],"study":[53],"asks":[54],"improves":[57],"performance":[58],"how":[60],"separate":[62],"task":[63],"from":[65,95],"presentation":[66],"statistics":[67,96],"inside":[68],"a":[69,88,132,153,192,216,223,244],"VLM\u2019s":[70],"latent":[71],"space.":[72],"Prior":[73],"work":[74],"probes":[75],"isolated":[76],"concepts":[77],"treats":[79],"algorithmically":[81],"(e.g.,":[82],"meta-optimization);":[83],"however,":[84],"it":[85,184],"seldom":[86],"provides":[87],"unified,":[89],"VLM-specific":[90],"account":[91,246],"that":[92,262],"disentangles":[93],"conditions.":[99],"propose":[101],"Semantics\u2013Statistics":[102],"Disentanglement":[103],"(S":[104],"2":[105,129],"D),":[106],"formalizing":[107],"projections":[108],"onto":[109],"semantic":[110,173,264],"versus":[111],"statistical":[112,268],"subspaces,":[113],"derive":[115],"two":[116,141],"hypotheses:":[117],"Semantic":[118],"Boost":[119],"(SB)":[120],"Textual":[122],"Dominance":[123],"Enhancement":[124],"(TDE).":[125],"implement":[127],"S":[128],"D":[130],"with":[131,136],"paired,":[133],"two-head":[134],"probe":[135],"randomized":[137],"controls":[138],"evaluate":[140],"LLaVA":[142],"variants":[143],"one-shot":[145],"examples":[147],"across":[148],"six":[149],"VQA":[150],"datasets":[151],"counterfactual-prompting":[154],"setup":[155],"Hateful":[157,209],"Memes.":[158],"Results":[159],"by":[160,227],"hypothesis":[161],"are":[162,238],"as":[163],"follows.":[164],"For":[165,198],"SB:":[166],"accuracy":[167],"gains":[168],"concentrate":[169],"where":[170,183],"zero-shot":[172,229,235],"signal":[174],"weak":[176],"become":[178],"muted":[179],"slightly":[181],"negative":[182],"already":[186],"strong,":[187],"representation":[189],"readouts":[190],"show":[191],"higher":[193],"semantic-over-statistical":[194],"ratio":[195],"after":[196],"ICL.":[197],"TDE:":[199],"counterfactual":[200],"prompting":[201],"raises":[202],"F1":[203],"lowers":[205],"cross-label":[206],"similarity":[207],"Memes,":[210],"indicating":[211],"reduced":[212],"label":[213],"bias":[214],"shift":[217,253],"toward":[218],"text-appropriate":[219],"dominance.":[220],"In":[221],"addition,":[222],"simple":[224],"selector":[225],"guided":[226],"state":[230],"outperforms":[231],"fixed":[232],"strategies":[233,261],"comparable.":[239],"Together,":[240],"these":[241],"findings":[242],"offer":[243],"principled":[245],"yield":[255],"actionable":[256],"diagnostics":[257],"for":[258],"choosing":[259],"context":[260],"maximize":[263],"gain":[265],"while":[266],"limiting":[267],"amplification.":[269]},"counts_by_year":[],"updated_date":"2026-11-12T12:28:57.059296","created_date":"2025-11-25T00:00:00"}