{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.001558899,0.0006177375,0.0005331583,0.002459415,0.0001201903,0.0004668691,0.0008805029,0.0003037974,0.0001363275],"category_scores_gemma":[0.0001145746,0.0004510593,0.00003554866,0.002245059,0.001720622,0.0004246929,0.0004687195,0.0008849122,0.0001655092],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003137655,"about_ca_system_score_gemma":0.001321081,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001259262,"about_ca_topic_score_gemma":0.003035399,"domain_scores_codex":[0.9957657,0.0001169831,0.0003393031,0.001677203,0.001227754,0.0008730626],"domain_scores_gemma":[0.9986338,0.00007333357,0.0002537931,0.000712248,0.00007351633,0.00025332],"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":[0.0007450302,0.0005611319,0.00002575036,0.001596967,0.0001654556,0.001710759,0.009431924,0.8288456,0.02716269,0.003209532,0.03654252,0.09000267],"study_design_scores_gemma":[0.001351775,0.0003457711,0.000004440612,0.001682129,0.00009272566,0.0001451794,0.0006715312,0.9921969,0.0002032242,0.001408846,0.000917664,0.0009798408],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.002106696,0.004394411,0.8979208,0.0002941325,0.001061272,0.004997605,0.0003565234,0.001903447,0.08696512],"genre_scores_gemma":[0.1841793,0.0000327193,0.649011,0.0001656316,0.001486727,0.0002329617,0.0001019005,0.004397617,0.1603921],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.2489097,"threshold_uncertainty_score":0.9997941,"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."}}