{"id":"https://openalex.org/W7138190816","doi":"https://doi.org/10.1609/aaai.v40i14.38121","title":"SRD: Reinforcement-Learned Semantic Perturbation for Backdoor Defense in VLMs","display_name":"SRD: Reinforcement-Learned Semantic Perturbation for Backdoor Defense in VLMs","publication_year":2026,"publication_date":"2026-03-14","ids":{"openalex":"https://openalex.org/W7138190816","doi":"https://doi.org/10.1609/aaai.v40i14.38121"},"language":"en","primary_location":{"id":"doi:10.1609/aaai.v40i14.38121","doi":"https://doi.org/10.1609/aaai.v40i14.38121","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i14.38121","pdf_url":null,"source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":["jufo-2","norway-1"],"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"volume":null,"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"diamond","oa_url":"https://doi.org/10.1609/aaai.v40i14.38121","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5123664594","display_name":"Shuhan Xu","orcid":"https://orcid.org/0009-0001-4378-5958","observed_orcids":["https://orcid.org/0009-0001-4378-5958"]},"institutions":[{"id":"https://openalex.org/I37461747","display_name":"Wuhan University","ror":"https://ror.org/033vjfk17","country_code":"CN","type":"education","lineage":["https://openalex.org/I37461747"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Shuhan Xu","raw_affiliation_strings":["Wuhan University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Wuhan University","institution_ids":["https://openalex.org/I37461747"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129719250","display_name":"Siyuan Liang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I172675005","display_name":"Nanyang Technological University","ror":"https://ror.org/02e7b5302","country_code":"SG","type":"education","lineage":["https://openalex.org/I172675005"]}],"countries":["SG"],"is_corresponding":false,"raw_author_name":"Siyuan Liang","raw_affiliation_strings":["Nanyang Technological University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Nanyang Technological University","institution_ids":["https://openalex.org/I172675005"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129743485","display_name":"Hongling Zheng","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I37461747","display_name":"Wuhan University","ror":"https://ror.org/033vjfk17","country_code":"CN","type":"education","lineage":["https://openalex.org/I37461747"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Hongling Zheng","raw_affiliation_strings":["Wuhan University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Wuhan University","institution_ids":["https://openalex.org/I37461747"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129699985","display_name":"Aishan Liu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I82880672","display_name":"Beihang University","ror":"https://ror.org/00wk2mp56","country_code":"CN","type":"education","lineage":["https://openalex.org/I82880672"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Aishan Liu","raw_affiliation_strings":["Beihang University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Beihang University","institution_ids":["https://openalex.org/I82880672"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5076370694","display_name":"Xinbiao Wang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I172675005","display_name":"Nanyang Technological University","ror":"https://ror.org/02e7b5302","country_code":"SG","type":"education","lineage":["https://openalex.org/I172675005"]}],"countries":["SG"],"is_corresponding":false,"raw_author_name":"Xinbiao Wang","raw_affiliation_strings":["Nanyang Technological University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Nanyang Technological University","institution_ids":["https://openalex.org/I172675005"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129651598","display_name":"Yong Luo","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I37461747","display_name":"Wuhan University","ror":"https://ror.org/033vjfk17","country_code":"CN","type":"education","lineage":["https://openalex.org/I37461747"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yong Luo","raw_affiliation_strings":["Wuhan University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Wuhan University","institution_ids":["https://openalex.org/I37461747"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129676112","display_name":"Fu Lin","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I37461747","display_name":"Wuhan University","ror":"https://ror.org/033vjfk17","country_code":"CN","type":"education","lineage":["https://openalex.org/I37461747"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Fu Lin","raw_affiliation_strings":["Wuhan University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Wuhan University","institution_ids":["https://openalex.org/I37461747"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128164600","display_name":"Leszek Rutkowski","orcid":"https://orcid.org/0000-0001-6960-9525","observed_orcids":["https://orcid.org/0000-0001-6960-9525"]},"institutions":[{"id":"https://openalex.org/I66083562","display_name":"Systems Research Institute","ror":"https://ror.org/0111cp837","country_code":"PL","type":"facility","lineage":["https://openalex.org/I66083562","https://openalex.org/I99542240"]},{"id":"https://openalex.org/I686019","display_name":"AGH University of Krakow","ror":"https://ror.org/00bas1c41","country_code":"PL","type":"education","lineage":["https://openalex.org/I686019"]},{"id":"https://openalex.org/I99542240","display_name":"Polish Academy of Sciences","ror":"https://ror.org/01dr6c206","country_code":"PL","type":"government","lineage":["https://openalex.org/I99542240"]}],"countries":["PL"],"is_corresponding":false,"raw_author_name":"Leszek Rutkowski","raw_affiliation_strings":["Systems Research Institute of the Polish Academy of Sciences\nAGH University of Krakow\nSAN University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Systems Research Institute of the Polish Academy of Sciences\nAGH University of Krakow\nSAN University","institution_ids":["https://openalex.org/I66083562","https://openalex.org/I686019","https://openalex.org/I99542240"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5074103823","display_name":"Dacheng Tao","orcid":"https://orcid.org/0000-0001-7225-5449","observed_orcids":["https://orcid.org/0000-0001-7225-5449","https://orcid.org/0000-0002-5979-578X","https://orcid.org/0000-0002-9209-036X"]},"institutions":[{"id":"https://openalex.org/I172675005","display_name":"Nanyang Technological University","ror":"https://ror.org/02e7b5302","country_code":"SG","type":"education","lineage":["https://openalex.org/I172675005"]}],"countries":["SG"],"is_corresponding":false,"raw_author_name":"Dacheng Tao","raw_affiliation_strings":["Nanyang Technological University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Nanyang Technological University","institution_ids":["https://openalex.org/I172675005"]}]}],"institutions":[],"countries_distinct_count":3,"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":0.12159048,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":"40","issue":"14","first_page":"11397","last_page":"11405"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.9994999766349792,"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":0.9994999766349792,"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/T12357","display_name":"Digital Media Forensic Detection","score":9.999999747378752e-05,"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/T11424","display_name":"Security and Verification in Computing","score":0.0,"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/backdoor-attacks","display_name":"backdoor attacks","score":0.8989999890327454},{"id":"https://openalex.org/keywords/vision-language-models","display_name":"vision-language models","score":0.9649999737739563},{"id":"https://openalex.org/keywords/backdoor-defense","display_name":"backdoor defense","score":0.847000002861023},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"reinforcement learning","score":0.7570000290870667},{"id":"https://openalex.org/keywords/semantic-perturbation","display_name":"semantic perturbation","score":0.8629999756813049},{"id":"https://openalex.org/keywords/image-captioning","display_name":"image captioning","score":0.75},{"id":"https://openalex.org/keywords/attention-manipulation","display_name":"attention manipulation","score":0.7229999899864197},{"id":"https://openalex.org/keywords/deep-q-network","display_name":"deep Q-network","score":0.6460000276565552},{"id":"https://openalex.org/keywords/adversarial-robustness","display_name":"adversarial robustness","score":0.6790000200271606}],"concepts":[{"id":"https://openalex.org/C2781045450","wikidata":"https://www.wikidata.org/wiki/Q254569","display_name":"Backdoor","level":2,"score":0.9890000224113464},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6492999792098999},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.5174000263214111},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4528000056743622},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.4390999972820282},{"id":"https://openalex.org/C2777530160","wikidata":"https://www.wikidata.org/wiki/Q41796","display_name":"Sentence","level":2,"score":0.3573000133037567},{"id":"https://openalex.org/C2776459999","wikidata":"https://www.wikidata.org/wiki/Q2119376","display_name":"Fidelity","level":2,"score":0.34529998898506165},{"id":"https://openalex.org/C2778180026","wikidata":"https://www.wikidata.org/wiki/Q18378163","display_name":"Semantic heterogeneity","level":4,"score":0.3343000113964081},{"id":"https://openalex.org/C157657479","wikidata":"https://www.wikidata.org/wiki/Q2367247","display_name":"Closed captioning","level":3,"score":0.32670000195503235},{"id":"https://openalex.org/C2777413886","wikidata":"https://www.wikidata.org/wiki/Q3276013","display_name":"Fluency","level":2,"score":0.32190001010894775},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.30169999599456787},{"id":"https://openalex.org/C115961682","wikidata":"https://www.wikidata.org/wiki/Q860623","display_name":"Image (mathematics)","level":2,"score":0.28870001435279846},{"id":"https://openalex.org/C204806902","wikidata":"https://www.wikidata.org/wiki/Q2333581","display_name":"Semantic security","level":5,"score":0.28679999709129333},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.28380000591278076},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.2718000113964081},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.2694000005722046},{"id":"https://openalex.org/C55439883","wikidata":"https://www.wikidata.org/wiki/Q360812","display_name":"Correctness","level":2,"score":0.26649999618530273},{"id":"https://openalex.org/C2776436953","wikidata":"https://www.wikidata.org/wiki/Q5163215","display_name":"Consistency (knowledge bases)","level":2,"score":0.2662999927997589},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.26170000433921814}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1609/aaai.v40i14.38121","doi":"https://doi.org/10.1609/aaai.v40i14.38121","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i14.38121","pdf_url":null,"source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":["jufo-2","norway-1"],"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"}],"best_oa_location":{"id":"doi:10.1609/aaai.v40i14.38121","doi":"https://doi.org/10.1609/aaai.v40i14.38121","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i14.38121","pdf_url":null,"source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":["jufo-2","norway-1"],"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"sustainable_development_goals":[],"sustainable_development_goals_aurora":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7160327541","https://openalex.org/W7202281374","https://openalex.org/W7153305462","https://openalex.org/W4417127580","https://openalex.org/W4416953404","https://openalex.org/W4404563144","https://openalex.org/W7137860081","https://openalex.org/W7161906593","https://openalex.org/W7129381619","https://openalex.org/W7106662527"],"abstract_inverted_index":{"Visual":[0],"language":[1],"models":[2],"(VLMs)":[3],"have":[4,15],"made":[5],"significant":[6],"progress":[7],"in":[8],"image":[9,136],"captioning":[10],"tasks,":[11],"yet":[12,187],"recent":[13],"studies":[14],"found":[16],"they":[17],"are":[18,42],"vulnerable":[19],"to":[20,45,51,127,131,144,183,209],"backdoor":[21,90,115,205],"attacks.":[22],"Attackers":[23],"can":[24],"inject":[25],"undetectable":[26],"perturbations":[27,130],"into":[28],"the":[29,52,58,75,85,149,170,177,181,222],"data":[30],"during":[31],"inference,":[32],"triggering":[33],"abnormal":[34,78],"behavior":[35,116],"and":[36,47,54,88,96,147,173,202,211],"generating":[37],"malicious":[38,152],"captions.":[39],"These":[40],"attacks":[41,91],"particularly":[43],"challenging":[44],"detect":[46],"defend":[48],"against":[49],"due":[50],"stealthiness":[53],"cross-modal":[55],"propagation":[56],"of":[57,84,122,135,151,176],"trigger":[59,123],"signals.":[60],"In":[61],"this":[62],"paper,":[63],"we":[64,103,158],"identify":[65],"two":[66],"key":[67],"vulnerabilities":[68],"by":[69],"analyzing":[70],"existing":[71],"attack":[72],"patterns:":[73],"(1)":[74],"model":[76],"exhibits":[77],"attention":[79,146],"concentration":[80],"on":[81,100,221],"certain":[82],"regions":[83,134],"input":[86],"image,":[87],"(2)":[89],"often":[92],"induce":[93],"semantic":[94,164,171],"drift":[95],"sentence":[97],"incoherence.":[98],"Based":[99],"these":[101],"insights,":[102],"propose":[104],"Semantic":[105],"Reward":[106],"Defense":[107],"(SRD),":[108],"a":[109,139,160,185,192],"reinforcement":[110],"learning":[111],"framework":[112],"that":[113,197],"mitigates":[114,199],"without":[117],"requiring":[118],"any":[119],"prior":[120],"knowledge":[121],"patterns.":[124],"SRD":[125,190],"learns":[126],"apply":[128],"discrete":[129],"sensitive":[132],"contextual":[133],"inputs":[137],"via":[138],"deep":[140],"Q-network":[141],"policy,":[142],"aiming":[143],"confuse":[145],"disrupt":[148],"activation":[150],"paths.":[153],"To":[154],"guide":[155],"policy":[156],"optimization,":[157],"design":[159],"reward":[161],"signal":[162],"named":[163],"fidelity":[165],"score,":[166],"which":[167],"jointly":[168],"assesses":[169],"consistency":[172],"linguistic":[174],"fluency":[175],"generated":[178],"captions,":[179],"encouraging":[180],"agent":[182],"achieve":[184],"robust":[186],"faithful":[188],"output.":[189],"offers":[191],"trigger-agnostic,":[193],"policy-interpretable":[194],"defense":[195],"paradigm":[196],"effectively":[198],"local":[200],"(TrojVLM)":[201],"global":[203],"(Shadowcast)":[204],"attacks,":[206],"reducing":[207],"ASR":[208],"3.6%":[210],"5.6%":[212],"respectively,":[213],"with":[214],"less":[215],"than":[216],"15%":[217],"average":[218],"CIDEr":[219],"drop":[220],"clean":[223],"inputs.":[224]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-03-18T00:00:00"}