{"id":"https://openalex.org/W7160297420","doi":"https://doi.org/10.1109/wacv61042.2026.00455","title":"Safe Vision-Language Models via Unsafe Weights Manipulation","display_name":"Safe Vision-Language Models via Unsafe Weights Manipulation","publication_year":2026,"publication_date":"2026-03-06","ids":{"openalex":"https://openalex.org/W7160297420","doi":"https://doi.org/10.1109/wacv61042.2026.00455"},"language":"en","primary_location":{"id":"doi:10.1109/wacv61042.2026.00455","is_oa":false,"landing_page_url":"https://doi.org/10.1109/wacv61042.2026.00455","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2026 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5113686107","display_name":"Moreno D\u2019Inc\u00e0","orcid":null},"institutions":[{"id":"https://openalex.org/I193223587","display_name":"University of Trento","ror":"https://ror.org/05trd4x28","country_code":"IT","type":"education","lineage":["https://openalex.org/I193223587"]}],"countries":["IT"],"is_corresponding":false,"raw_author_name":"Moreno D\u2019Inc\u00e0","raw_affiliation_strings":["University of Trento"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Trento","institution_ids":["https://openalex.org/I193223587"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5004548591","display_name":"Elia Peruzzo","orcid":"https://orcid.org/0000-0003-0119-3783"},"institutions":[{"id":"https://openalex.org/I193223587","display_name":"University of Trento","ror":"https://ror.org/05trd4x28","country_code":"IT","type":"education","lineage":["https://openalex.org/I193223587"]}],"countries":["IT"],"is_corresponding":false,"raw_author_name":"Elia Peruzzo","raw_affiliation_strings":["University of Trento"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Trento","institution_ids":["https://openalex.org/I193223587"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135388997","display_name":"Xingqian Xu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xingqian Xu","raw_affiliation_strings":["NVIDIA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"NVIDIA","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5002072267","display_name":"Humphrey Shi","orcid":"https://orcid.org/0000-0002-2922-5663"},"institutions":[{"id":"https://openalex.org/I130701444","display_name":"Georgia Institute of Technology","ror":"https://ror.org/01zkghx44","country_code":"US","type":"education","lineage":["https://openalex.org/I130701444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Humphrey Shi","raw_affiliation_strings":["Georgia Tech"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Georgia Tech","institution_ids":["https://openalex.org/I130701444"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135303217","display_name":"Nicu Sebe","orcid":null},"institutions":[{"id":"https://openalex.org/I193223587","display_name":"University of Trento","ror":"https://ror.org/05trd4x28","country_code":"IT","type":"education","lineage":["https://openalex.org/I193223587"]}],"countries":["IT"],"is_corresponding":false,"raw_author_name":"Nicu Sebe","raw_affiliation_strings":["University of Trento"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Trento","institution_ids":["https://openalex.org/I193223587"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5135340426","display_name":"Massimiliano Mancini","orcid":null},"institutions":[{"id":"https://openalex.org/I193223587","display_name":"University of Trento","ror":"https://ror.org/05trd4x28","country_code":"IT","type":"education","lineage":["https://openalex.org/I193223587"]}],"countries":["IT"],"is_corresponding":false,"raw_author_name":"Massimiliano Mancini","raw_affiliation_strings":["University of Trento"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Trento","institution_ids":["https://openalex.org/I193223587"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":6,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.68915605,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"4682","last_page":"4692"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.4352000057697296,"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":0.4352000057697296,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.0746999979019165,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.07320000231266022,"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/set","display_name":"Set (abstract data type)","score":0.2632000148296356},{"id":"https://openalex.org/keywords/identification","display_name":"Identification (biology)","score":0.25589999556541443},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.2549999952316284},{"id":"https://openalex.org/keywords/component","display_name":"Component (thermodynamics)","score":0.2531999945640564},{"id":"https://openalex.org/keywords/stability","display_name":"Stability (learning theory)","score":0.24250000715255737}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5490000247955322},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.2632000148296356},{"id":"https://openalex.org/C116834253","wikidata":"https://www.wikidata.org/wiki/Q2039217","display_name":"Identification (biology)","level":2,"score":0.25589999556541443},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.2549999952316284},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.2531999945640564},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.24500000476837158},{"id":"https://openalex.org/C112972136","wikidata":"https://www.wikidata.org/wiki/Q7595718","display_name":"Stability (learning theory)","level":2,"score":0.24250000715255737},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.23890000581741333},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.23770000040531158},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.23330000042915344}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.1109/wacv61042.2026.00455","is_oa":false,"landing_page_url":"https://doi.org/10.1109/wacv61042.2026.00455","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2026 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)","raw_type":"proceedings-article"},{"id":"pmh:oai:iris.unitn.it:11572/486932","is_oa":false,"landing_page_url":"https://hdl.handle.net/11572/486932","pdf_url":null,"source":{"id":"https://openalex.org/S4306401913","display_name":"Institutional Research Information System (Universit\u00e0 degli Studi di Trento)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I193223587","host_organization_name":"University of Trento","host_organization_lineage":["https://openalex.org/I193223587"],"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":"info:eu-repo/semantics/conferenceObject"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":35,"referenced_works":["https://openalex.org/W12634471","https://openalex.org/W1977295328","https://openalex.org/W2017814585","https://openalex.org/W2047643928","https://openalex.org/W2101105183","https://openalex.org/W2108598243","https://openalex.org/W2138011018","https://openalex.org/W2166049352","https://openalex.org/W2473930607","https://openalex.org/W2533598788","https://openalex.org/W2964194231","https://openalex.org/W3037492894","https://openalex.org/W3177096435","https://openalex.org/W4283172096","https://openalex.org/W4290943938","https://openalex.org/W4312261477","https://openalex.org/W4385894687","https://openalex.org/W4386066536","https://openalex.org/W4390871671","https://openalex.org/W4390871724","https://openalex.org/W4390873312","https://openalex.org/W4394593019","https://openalex.org/W4399528455","https://openalex.org/W4401044036","https://openalex.org/W4402670570","https://openalex.org/W4402704640","https://openalex.org/W4402716027","https://openalex.org/W4402959574","https://openalex.org/W4403195962","https://openalex.org/W4403842422","https://openalex.org/W4404781272","https://openalex.org/W4404879652","https://openalex.org/W4405181600","https://openalex.org/W4412886755","https://openalex.org/W7133193597"],"related_works":[],"abstract_inverted_index":{"Vision-language":[0],"models":[1],"(VLMs)":[2],"often":[3],"inherit":[4],"the":[5,28,77,129,135,149],"biases":[6],"and":[7,92,117,125,154,176],"unsafe":[8,20,32,118,126,161,175],"associations":[9],"present":[10],"within":[11],"their":[12,22],"large-scale":[13],"training":[14],"dataset.":[15],"While":[16],"recent":[17],"approaches":[18],"mitigate":[19],"behaviors,":[21],"evaluation":[23,47],"focuses":[24],"on":[25,31,37,81,160,169],"how":[26],"safe":[27,38,80,82,116,124,170],"model":[29,78,101],"is":[30,96],"inputs,":[33],"ignoring":[34],"potential":[35],"shortcomings":[36],"ones.":[39],"In":[40],"this":[41,65,85],"paper,":[42],"we":[43,67,87],"first":[44],"revise":[45],"safety":[46,58,153],"by":[48],"introducing":[49,105],"SafeGround,":[50],"a":[51,69,89,100,112],"new":[52],"set":[53,114],"of":[54,62,72,115],"metrics":[55],"that":[56,146,179],"evaluate":[57],"at":[59],"different":[60,90],"levels":[61],"granularity.":[63],"With":[64],"metric,":[66],"uncover":[68],"surprising":[70],"issue":[71],"training-based":[73,166],"methods:":[74],"they":[75],"make":[76,99],"less":[79],"inputs.":[83],"From":[84],"finding,":[86],"take":[88],"direction":[91],"explore":[93],"whether":[94],"it":[95],"possible":[97],"to":[98,120],"safer":[102],"without":[103],"training,":[104],"Unsafe":[106],"Weights":[107],"Manipulation":[108],"(UWM).":[109],"UWM":[110,147],"uses":[111],"calibration":[113],"instances":[119],"compare":[121],"activations":[122],"between":[123,152],"content,":[127],"identifying":[128],"most":[130],"important":[131],"parameters":[132],"for":[133],"processing":[134],"latter.":[136],"Their":[137],"values":[138],"are":[139],"then":[140],"manipulated":[141],"via":[142],"negation.":[143],"Experiments":[144],"show":[145],"achieves":[148],"best":[150],"tradeoff":[151],"knowledge":[155],"preservation,":[156],"consistently":[157],"improving":[158],"VLMs":[159],"queries":[162],"while":[163],"outperforming":[164],"even":[165],"state-of-the-art":[167],"methods":[168],"ones.Warning:":[171],"This":[172],"paper":[173],"includes":[174],"harmful":[177],"content":[178,184],"may":[180],"be":[181],"disturbing.":[182],"Such":[183],"has":[185],"been":[186],"blurred.":[187]},"counts_by_year":[],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2026-05-06T00:00:00"}