{"id":"https://openalex.org/W7160199243","doi":"https://doi.org/10.48550/arxiv.2605.00326","title":"Prompt-Induced Score Variance in Zero-Shot Binary Vision-Language Safety Classification","display_name":"Prompt-Induced Score Variance in Zero-Shot Binary Vision-Language Safety Classification","publication_year":2026,"publication_date":"2026-05-01","ids":{"openalex":"https://openalex.org/W7160199243","doi":"https://doi.org/10.48550/arxiv.2605.00326"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.00326","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.00326","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":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.00326","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5120317474","display_name":"Charles Weng","orcid":null},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Weng, Charles","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5039145854","display_name":"Dingwen Li","orcid":"https://orcid.org/0000-0002-9231-7317"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Dingwen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5135098475","display_name":"Alexander Martin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Martin, Alexander","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":3,"corresponding_author_ids":["https://openalex.org/A5120317474"],"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.6620000004768372,"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"}},"topics":[{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.6620000004768372,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.05959999933838844,"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/T10809","display_name":"Occupational Health and Safety Research","score":0.04170000180602074,"subfield":{"id":"https://openalex.org/subfields/3614","display_name":"Radiological and Ultrasound Technology"},"field":{"id":"https://openalex.org/fields/36","display_name":"Health Professions"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/reliability","display_name":"Reliability (semiconductor)","score":0.5906999707221985},{"id":"https://openalex.org/keywords/ranking","display_name":"Ranking (information retrieval)","score":0.5722000002861023},{"id":"https://openalex.org/keywords/variance","display_name":"Variance (accounting)","score":0.555400013923645},{"id":"https://openalex.org/keywords/frame","display_name":"Frame (networking)","score":0.4611000120639801},{"id":"https://openalex.org/keywords/isotonic-regression","display_name":"Isotonic regression","score":0.44359999895095825},{"id":"https://openalex.org/keywords/binary-number","display_name":"Binary number","score":0.43290001153945923},{"id":"https://openalex.org/keywords/calibration","display_name":"Calibration","score":0.4235999882221222},{"id":"https://openalex.org/keywords/binary-classification","display_name":"Binary classification","score":0.3668999969959259}],"concepts":[{"id":"https://openalex.org/C105795698","wikidata":"https://www.wikidata.org/wiki/Q12483","display_name":"Statistics","level":1,"score":0.6427000164985657},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.6328999996185303},{"id":"https://openalex.org/C43214815","wikidata":"https://www.wikidata.org/wiki/Q7310987","display_name":"Reliability (semiconductor)","level":3,"score":0.5906999707221985},{"id":"https://openalex.org/C189430467","wikidata":"https://www.wikidata.org/wiki/Q7293293","display_name":"Ranking (information retrieval)","level":2,"score":0.5722000002861023},{"id":"https://openalex.org/C196083921","wikidata":"https://www.wikidata.org/wiki/Q7915758","display_name":"Variance (accounting)","level":2,"score":0.555400013923645},{"id":"https://openalex.org/C126042441","wikidata":"https://www.wikidata.org/wiki/Q1324888","display_name":"Frame (networking)","level":2,"score":0.4611000120639801},{"id":"https://openalex.org/C17418463","wikidata":"https://www.wikidata.org/wiki/Q3455874","display_name":"Isotonic regression","level":3,"score":0.44359999895095825},{"id":"https://openalex.org/C48372109","wikidata":"https://www.wikidata.org/wiki/Q3913","display_name":"Binary number","level":2,"score":0.43290001153945923},{"id":"https://openalex.org/C165838908","wikidata":"https://www.wikidata.org/wiki/Q736777","display_name":"Calibration","level":2,"score":0.4235999882221222},{"id":"https://openalex.org/C66905080","wikidata":"https://www.wikidata.org/wiki/Q17005494","display_name":"Binary classification","level":3,"score":0.3668999969959259},{"id":"https://openalex.org/C83546350","wikidata":"https://www.wikidata.org/wiki/Q1139051","display_name":"Regression","level":2,"score":0.32409998774528503},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.322299987077713},{"id":"https://openalex.org/C110121322","wikidata":"https://www.wikidata.org/wiki/Q865811","display_name":"Distribution (mathematics)","level":2,"score":0.3174000084400177},{"id":"https://openalex.org/C35405484","wikidata":"https://www.wikidata.org/wiki/Q4967066","display_name":"Brier score","level":2,"score":0.3154999911785126},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.31360000371932983},{"id":"https://openalex.org/C22679943","wikidata":"https://www.wikidata.org/wiki/Q159375","display_name":"Standard deviation","level":2,"score":0.3001999855041504},{"id":"https://openalex.org/C4679612","wikidata":"https://www.wikidata.org/wiki/Q866298","display_name":"Aggregate (composite)","level":2,"score":0.29750001430511475},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.29750001430511475},{"id":"https://openalex.org/C149782125","wikidata":"https://www.wikidata.org/wiki/Q160039","display_name":"Econometrics","level":1,"score":0.27459999918937683},{"id":"https://openalex.org/C48921125","wikidata":"https://www.wikidata.org/wiki/Q10861030","display_name":"Linear regression","level":2,"score":0.27059999108314514},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.26750001311302185},{"id":"https://openalex.org/C152877465","wikidata":"https://www.wikidata.org/wiki/Q208042","display_name":"Regression analysis","level":2,"score":0.2563999891281128},{"id":"https://openalex.org/C149441793","wikidata":"https://www.wikidata.org/wiki/Q200726","display_name":"Probability distribution","level":2,"score":0.25040000677108765}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.00326","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.00326","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":"article"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.00326","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.00326","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":"article"},"sustainable_development_goals":[{"score":0.5615237355232239,"display_name":"Peace, Justice and strong institutions","id":"https://metadata.un.org/sdg/16"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Single-prompt":[0],"first-token":[1,184],"probabilities":[2,45],"from":[3],"zero-shot":[4,182],"vision-language":[5],"model":[6],"(VLM)":[7],"safety":[8,52,185],"classifiers":[9],"are":[10,19,119,156],"treated":[11],"as":[12,161,176,194],"decision":[13],"scores,":[14],"but":[15],"we":[16],"show":[17],"they":[18],"unreliable":[20],"under":[21],"semantically":[22],"equivalent":[23,38],"prompt":[24,159],"reformulation:":[25],"even":[26],"when":[27,154],"the":[28,47,114,122,132,149],"binary":[29],"label":[30],"is":[31,60],"constrained":[32],"to":[33,92,113],"a":[34,71,93,162,169,177,195],"fixed":[35],"output":[36],"position,":[37],"prompts":[39],"can":[40],"induce":[41],"materially":[42],"different":[43],"unsafe":[44],"for":[46,171,181],"same":[48,115],"sample.":[49],"Across":[50],"multimodal":[51],"benchmarks":[53],"and":[54,66,87,97,109,128,130,187],"multiple":[55],"VLM":[56,183],"families,":[57],"cross-prompt":[58],"variance":[59],"strongly":[61],"associated":[62],"with":[63,191],"prompt-level":[64],"disagreement":[65],"higher":[67],"error,":[68],"making":[69],"it":[70],"useful":[72],"fragility":[73],"diagnostic.":[74],"A":[75],"training-free":[76],"mean":[77,150,192],"ensemble":[78],"improves":[79],"NLL":[80,101],"on":[81,89,125,138,142,146],"all":[82],"14":[83],"dataset-model":[84],"evaluation":[85,190],"pairs":[86],"ECE":[88],"12/14":[90],"relative":[91],"train-selected":[94,123],"single-prompt":[95],"baseline,":[96],"wins":[98],"more":[99],"head-to-head":[100],"comparisons":[102],"than":[103,168],"labeled":[104],"temperature":[105],"scaling,":[106,108],"Platt":[107],"isotonic":[110],"regression":[111],"applied":[112],"prompt.":[116],"Ranking":[117],"gains":[118,153],"consistent":[120,137],"against":[121,131],"baseline":[124],"both":[126],"AUROC":[127],"AUPRC,":[129],"full":[133],"15-prompt":[134],"distribution":[135],"remain":[136],"AUPRC":[139],"while":[140],"softening":[141],"AUROC.":[143],"Labeled":[144],"calibration":[145],"top":[147],"of":[148],"provides":[151],"further":[152],"labels":[155],"available,":[157],"identifying":[158],"averaging":[160],"strong":[163],"label-free":[164,197],"first":[165],"stage":[166],"rather":[167],"replacement":[170],"calibration.":[172],"We":[173],"frame":[174],"this":[175],"reliability":[178,198],"stress":[179],"test":[180],"scores":[186],"recommend":[188],"prompt-family":[189],"aggregation":[193],"standard":[196],"baseline.":[199]},"counts_by_year":[],"updated_date":"2026-05-05T06:12:25.323381","created_date":"2026-05-05T00:00:00"}