{"id":"W4409361716","doi":"10.1609/aaai.v39i26.34954","title":"Internal Activation Revision: Safeguarding Vision Language Models Without Parameter Update","year":2025,"lang":"en","type":"article","venue":"Proceedings of the AAAI Conference on Artificial Intelligence","topic":"Adversarial Robustness in Machine Learning","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Safeguarding; Computer science; Linguistics; Natural language processing; Philosophy; Medicine","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":[0.003067954,0.001446833,0.0008800921,0.000490414,0.0004891784,0.001416425,0.002932031,0.001430665,0.002777761],"category_scores_gemma":[0.01944879,0.0005792219,0.0008719039,0.0002180779,0.001654441,0.002955446,0.002947614,0.002931828,0.001121499],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007870369,"about_ca_system_score_gemma":0.00134431,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002337816,"about_ca_topic_score_gemma":0.002011207,"domain_scores_codex":[0.9980752,0.0006644094,0.000100608,0.0003929727,0.0005303019,0.0002364221],"domain_scores_gemma":[0.993159,0.002846095,0.0006450941,0.002336916,0.0008112434,0.0002016567],"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":[0.0010859,0.0002121947,0.005355862,0.0002267786,0.0001500907,0.0006213098,0.00102404,0.5122783,0.05215767,0.03076127,0.008852791,0.3872739],"study_design_scores_gemma":[0.00002399507,0.00009936427,0.0001733102,0.00001484289,0.00002175987,0.00007973431,0.00004132791,0.9720129,0.0155871,0.01064026,0.001287178,0.00001814625],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.08309466,0.0002535483,0.9027569,0.0006684785,0.0001316543,0.000204379,0.000126168,0.009302385,0.003461806],"genre_scores_gemma":[0.8992461,0.00008540506,0.09607854,0.0005109033,0.00006625339,0.0001446133,0.0001850456,0.0006582353,0.003024904],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.003067954,"threshold_uncertainty_score":0.0162251,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03992248449768911,"score_gpt":0.3291544716685126,"score_spread":0.2892319871708235,"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."}}