{"id":"W4404780969","doi":"10.18653/v1/2024.findings-emnlp.990","title":"Improving Adversarial Robustness in Vision-Language Models with Architecture and Prompt Design","year":2024,"lang":"en","type":"article","venue":"","topic":"Adversarial Robustness in Machine Learning","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":false,"ca_institutions":"","funders":"Alliance de recherche numérique du Canada; Canada Excellence Research Chairs, Government of Canada; Canadian Institute for Advanced Research","keywords":"Robustness (evolution); Computer science; Adversarial system; Architecture; Artificial intelligence; Computer architecture","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0005901291,0.000232249,0.0002106892,0.0003002963,0.00009852203,0.0004592169,0.0005284717,0.0000954313,0.00001491503],"category_scores_gemma":[0.00005274114,0.0001673895,0.00002955618,0.0006277584,0.00006725923,0.001151155,0.0004134155,0.0005300598,0.000003406222],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006068588,"about_ca_system_score_gemma":0.0001656921,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002249553,"about_ca_topic_score_gemma":0.00006323671,"domain_scores_codex":[0.9982246,0.0001580916,0.0002054538,0.0007244471,0.0003186988,0.0003687629],"domain_scores_gemma":[0.9991692,0.0002962525,0.00004135465,0.0003757806,0.00002538684,0.00009204687],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00003772115,0.00001517862,0.0000190888,0.00005840559,0.00001139731,0.000267111,0.002784354,0.8835185,0.000345408,0.01379226,0.00001966828,0.09913096],"study_design_scores_gemma":[0.0005020185,0.0001289366,0.00004285455,0.0001335476,0.000008141068,0.00008916549,0.0001381697,0.9964595,0.00014785,0.002073238,0.00002138644,0.000255173],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.008891932,0.0003089839,0.9884343,0.0005906523,0.000241932,0.0003771872,5.396287e-7,0.0004536395,0.0007008946],"genre_scores_gemma":[0.5638274,0.000002027707,0.4358442,0.00004318087,0.00007685124,0.00001598441,9.004722e-7,0.00001983452,0.000169661],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.5549355,"threshold_uncertainty_score":0.6825945,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.008985468768117039,"score_gpt":0.2443913261637576,"score_spread":0.2354058573956406,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}