{"id":"https://openalex.org/W4403586310","doi":"https://doi.org/10.48550/arxiv.2409.04142","title":"Context is the Key: Backdoor Attacks for In-Context Learning with Vision Transformers","display_name":"Context is the Key: Backdoor Attacks for In-Context Learning with Vision Transformers","publication_year":2024,"publication_date":"2024-09-06","ids":{"openalex":"https://openalex.org/W4403586310","doi":"https://doi.org/10.48550/arxiv.2409.04142"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2409.04142","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2409.04142","pdf_url":"https://arxiv.org/pdf/2409.04142","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2409.04142","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5043249066","display_name":"Gorka Abad","orcid":"https://orcid.org/0000-0002-6735-3623","observed_orcids":["https://orcid.org/0000-0002-6735-3623"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Abad, Gorka","raw_affiliation_strings":["Radboud University Ikerlan Research Centre"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Radboud University Ikerlan Research Centre","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5024072796","display_name":"Stjepan Picek","orcid":"https://orcid.org/0000-0001-7509-4337","observed_orcids":["https://orcid.org/0000-0001-7509-4337","https://orcid.org/0000-0002-9290-1580"]},"institutions":[{"id":"https://openalex.org/I145872427","display_name":"Radboud University Nijmegen","ror":"https://ror.org/016xsfp80","country_code":"NL","type":"education","lineage":["https://openalex.org/I145872427"]}],"countries":["NL"],"is_corresponding":false,"raw_author_name":"Picek, Stjepan","raw_affiliation_strings":["Radboud University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Radboud University","institution_ids":["https://openalex.org/I145872427"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5036908366","display_name":"Lorenzo Cavallaro","orcid":"https://orcid.org/0000-0002-3878-2680","observed_orcids":["https://orcid.org/0000-0002-3878-2680"]},"institutions":[{"id":"https://openalex.org/I45129253","display_name":"University College London","ror":"https://ror.org/02jx3x895","country_code":"GB","type":"education","lineage":["https://openalex.org/I124357947","https://openalex.org/I45129253"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Cavallaro, Lorenzo","raw_affiliation_strings":["University College London"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University College London","institution_ids":["https://openalex.org/I45129253"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5027736689","display_name":"Aitor Urbieta","orcid":"https://orcid.org/0000-0001-5836-4198","observed_orcids":["https://orcid.org/0000-0001-5836-4198"]},"institutions":[{"id":"https://openalex.org/I4210140267","display_name":"Ikerlan","ror":"https://ror.org/03hp1m080","country_code":"ES","type":"company","lineage":["https://openalex.org/I4210140267"]}],"countries":["ES"],"is_corresponding":false,"raw_author_name":"Urbieta, Aitor","raw_affiliation_strings":["Ikerlan Research Centre"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Ikerlan Research Centre","institution_ids":["https://openalex.org/I4210140267"]}]}],"institutions":[],"countries_distinct_count":3,"institutions_distinct_count":3,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":1,"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":1.0,"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"}},"topics":[{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":1.0,"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"}},{"id":"https://openalex.org/T12357","display_name":"Digital Media Forensic Detection","score":0.0,"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/1708","display_name":"Hardware and Architecture"},"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":0.9980000257492065},{"id":"https://openalex.org/keywords/in-context-learning","display_name":"in-context learning","score":0.9929999709129333},{"id":"https://openalex.org/keywords/vision-transformer","display_name":"vision transformer","score":0.9919999837875366},{"id":"https://openalex.org/keywords/data-poisoning-attacks","display_name":"data poisoning","score":0.9769999980926514},{"id":"https://openalex.org/keywords/prompt-based-attacks","display_name":"prompt-based attacks","score":0.515999972820282},{"id":"https://openalex.org/keywords/generalization-to-unseen-tasks","display_name":"generalization to unseen tasks","score":0.6159999966621399},{"id":"https://openalex.org/keywords/prompt-robustness","display_name":"prompt robustness","score":0.7549999952316284}],"concepts":[{"id":"https://openalex.org/C2781045450","wikidata":"https://www.wikidata.org/wiki/Q254569","display_name":"Backdoor","level":2,"score":0.9850544333457947},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.5869985222816467},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.4979078769683838},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.48067912459373474},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4598672091960907},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.3670792877674103},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.36371445655822754},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.2213188111782074},{"id":"https://openalex.org/C95457728","wikidata":"https://www.wikidata.org/wiki/Q309","display_name":"History","level":0,"score":0.131289541721344},{"id":"https://openalex.org/C119599485","wikidata":"https://www.wikidata.org/wiki/Q43035","display_name":"Electrical engineering","level":1,"score":0.11209309101104736},{"id":"https://openalex.org/C166957645","wikidata":"https://www.wikidata.org/wiki/Q23498","display_name":"Archaeology","level":1,"score":0.0},{"id":"https://openalex.org/C165801399","wikidata":"https://www.wikidata.org/wiki/Q25428","display_name":"Voltage","level":2,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2409.04142","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2409.04142","pdf_url":"https://arxiv.org/pdf/2409.04142","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2409.04142","doi":"https://doi.org/10.48550/arxiv.2409.04142","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2409.04142","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2409.04142","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2409.04142","pdf_url":"https://arxiv.org/pdf/2409.04142","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4403586310.pdf","grobid_xml":"https://content.openalex.org/works/W4403586310.grobid-xml"},"referenced_works_count":17,"referenced_works":["https://openalex.org/W2798812533","https://openalex.org/W2887817889","https://openalex.org/W2896457183","https://openalex.org/W2963038646","https://openalex.org/W2965373594","https://openalex.org/W3094502228","https://openalex.org/W3130788031","https://openalex.org/W3199245537","https://openalex.org/W4226512186","https://openalex.org/W4319323646","https://openalex.org/W4385374045","https://openalex.org/W4386076112","https://openalex.org/W4387596607","https://openalex.org/W4390873795","https://openalex.org/W4394839259","https://openalex.org/W6884882052","https://openalex.org/W7133242526"],"related_works":["https://openalex.org/W4398157063","https://openalex.org/W7153305462","https://openalex.org/W4402753692","https://openalex.org/W4403883324","https://openalex.org/W4404782283","https://openalex.org/W4386075825","https://openalex.org/W7214753815","https://openalex.org/W4386066003","https://openalex.org/W4405468480","https://openalex.org/W4416425925"],"abstract_inverted_index":{"Due":[0],"to":[1,22,33,72,99,142,157,217],"the":[2,29,39,66,78,93,103,112,115,118,131,146,152,155,165,185,195,198,209,213],"high":[3],"cost":[4],"of":[5,31,68,114,179,187],"training,":[6],"large":[7],"model":[8],"(LM)":[9],"practitioners":[10],"commonly":[11],"use":[12],"pretrained":[13],"models":[14,58],"downloaded":[15],"from":[16,197,215],"untrusted":[17],"sources,":[18],"which":[19],"could":[20],"lead":[21],"owning":[23],"compromised":[24,107],"models.":[25],"In-context":[26],"learning":[27],"is":[28,106,124],"ability":[30,67],"LMs":[32],"perform":[34,73],"multiple":[35],"tasks":[36,75,162],"depending":[37,55,76],"on":[38,56,77,145,172],"prompt":[40],"or":[41],"context.":[42],"This":[43],"can":[44],"enable":[45],"new":[46,87],"attacks,":[47],"such":[48],"as":[49,191],"backdoor":[50,156],"attacks":[51],"with":[52,130],"dynamic":[53],"behavior":[54],"how":[57],"are":[59],"prompted.":[60],"In":[61],"this":[62],"paper,":[63],"we":[64,84,183],"leverage":[65],"vision":[69],"transformers":[70],"(ViTs)":[71],"different":[74],"prompts.":[79],"Then,":[80],"through":[81],"data":[82],"poisoning,":[83],"investigate":[85,184],"two":[86],"threats:":[88],"i)":[89],"task-specific":[90],"backdoors":[91,196],"where":[92],"attacker":[94],"chooses":[95],"a":[96,177],"target":[97,147],"task":[98,105,123],"attack,":[100,153],"and":[101,189],"only":[102],"selected":[104],"at":[108],"test":[109],"time":[110],"under":[111],"presence":[113],"trigger.":[116,132],"At":[117],"same":[119],"time,":[120],"any":[121],"other":[122],"not":[125],"affected,":[126],"even":[127,161],"if":[128],"prompted":[129],"We":[133,150,200],"succeeded":[134],"in":[135,208],"attacking":[136],"every":[137,173],"tested":[138,174],"model,":[139,175],"achieving":[140,176],"up":[141],"89.90\\%":[143,216],"degradation":[144,214],"task.":[148],"ii)":[149],"generalize":[151],"allowing":[154],"affect":[158],"\\emph{any}":[159],"task,":[160],"unseen":[163],"during":[164],"training":[166],"phase.":[167],"Our":[168],"attack":[169],"was":[170],"successful":[171],"maximum":[178],"$13\\times$":[180],"degradation.":[181],"Finally,":[182],"robustness":[186],"prompts":[188],"fine-tuning":[190],"techniques":[192],"for":[193],"removing":[194],"model.":[199],"found":[201],"that":[202],"these":[203],"methods":[204],"fall":[205],"short":[206],"and,":[207],"best":[210],"case,":[211],"reduce":[212],"73.46\\%.":[218]},"counts_by_year":[{"year":2025,"cited_by_count":1}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2024-10-21T00:00:00"}