{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0007728989,0.0002505512,0.0003128407,0.0002339779,0.0002612532,0.0004049216,0.002253308,0.0001114392,0.00003935922],"category_scores_gemma":[0.0008413835,0.0001888683,0.0001398655,0.0008103693,0.0001674949,0.001149951,0.0009310727,0.0005495775,0.00003469087],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009073546,"about_ca_system_score_gemma":0.00008198298,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000491598,"about_ca_topic_score_gemma":0.000001793886,"domain_scores_codex":[0.9979327,0.00004759468,0.0005850796,0.0005959807,0.0005235858,0.000315034],"domain_scores_gemma":[0.9984367,0.0001820995,0.0004479171,0.000422549,0.0004537029,0.00005696223],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00009691803,0.00005539988,0.0001424412,0.00004597182,0.00002078267,4.434816e-7,0.001472901,0.003655428,0.01761141,0.7799342,0.0001487287,0.1968154],"study_design_scores_gemma":[0.00003997161,0.00006424404,0.00006308701,0.0009053696,0.00001266822,0.000001487125,0.0003945306,0.6252788,0.2082788,0.1646782,0.0001064901,0.0001764167],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.08492848,0.00001992632,0.8952373,0.004882797,0.0006540399,0.0003556432,0.000001038933,0.0001386773,0.01378209],"genre_scores_gemma":[0.9793122,0.00002034938,0.01975944,0.0004523207,0.00006705702,0.00001065907,5.605995e-7,0.00001207308,0.0003653625],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8943837,"threshold_uncertainty_score":0.7701827,"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."}}