{"id":"W6891703365","doi":"10.48448/rf80-vf95","title":"Improving Adversarial Robustness in Vision-Language Models with Architecture and Prompt Design","year":2024,"lang":"en","type":"other","venue":"Underline Science Inc.","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Mila - Quebec Artificial Intelligence Institute","funders":"","keywords":"Robustness (evolution); Adversarial system; Software deployment; Architecture; Threat model","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002226489,0.001104024,0.0006008149,0.000398327,0.0003908584,0.001364318,0.001319885,0.001420631,0.00382857],"category_scores_gemma":[0.01212548,0.0004975568,0.0006104588,0.0001977574,0.001387818,0.002445218,0.002631435,0.002603482,0.001207244],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001041786,"about_ca_system_score_gemma":0.001490425,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001758668,"about_ca_topic_score_gemma":0.0028896,"domain_scores_codex":[0.9988274,0.0004829481,0.00005126764,0.0002369349,0.0002713597,0.0001300482],"domain_scores_gemma":[0.9960468,0.002243899,0.0003807615,0.0007053062,0.0004444825,0.0001787453],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0002263844,0.0001448161,0.001410747,0.0001813858,0.00006462578,0.0001841365,0.0001732959,0.8488713,0.01999106,0.04804129,0.003108344,0.07760269],"study_design_scores_gemma":[0.00001483234,0.00006887042,0.00007665106,0.00001570738,0.00001081198,0.000047189,0.00001823708,0.9762714,0.004643037,0.01759693,0.001223775,0.00001259961],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"other","genre_scores_codex":[0.02768947,0.0004457507,0.9641241,0.001026251,0.0001177946,0.00007921823,0.00007654107,0.002043061,0.004397747],"genre_scores_gemma":[0.8087393,0.0003955735,0.1847369,0.0006740023,0.00007475492,0.0001489822,0.0001698045,0.0003999998,0.004660724],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.00382857,"threshold_uncertainty_score":0.01280779,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01538263378773222,"score_gpt":0.2735358688257843,"score_spread":0.2581532350380521,"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."}}