{"id":"W4415549799","doi":"10.1007/978-3-032-08707-2_14","title":"Solution-Aware Vs Global ReLU Selection: Partial MILP Strikes Back for DNN Verification","year":2025,"lang":"en","type":"book-chapter","venue":"Lecture notes in computer science","topic":"Adversarial Robustness in Machine Learning","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Canada Research Chairs; University of Toronto","funders":"Agence Nationale de la Recherche","keywords":"Binary number; Set (abstract data type); Lipschitz continuity; Binary decision diagram; Upper and lower bounds; Branching (polymer chemistry); Branch and bound; Deep neural networks","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":[0.00357602,0.001687053,0.001405738,0.0005218606,0.0008611181,0.001846076,0.002060192,0.00210705,0.01114715],"category_scores_gemma":[0.01155345,0.0007481224,0.001092484,0.0004209394,0.001836532,0.003301336,0.003578954,0.003747586,0.001472019],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001099111,"about_ca_system_score_gemma":0.002807761,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001831833,"about_ca_topic_score_gemma":0.004477764,"domain_scores_codex":[0.9979672,0.0008642056,0.00007825741,0.0003354857,0.0004951465,0.0002596179],"domain_scores_gemma":[0.9956965,0.002556933,0.0001628835,0.001011465,0.0004106406,0.0001614997],"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":[0.0005010713,0.0001349023,0.0006132321,0.0002741228,0.0001174844,0.0001867217,0.0001001404,0.692955,0.005322635,0.1013226,0.01166746,0.1868046],"study_design_scores_gemma":[0.00002111835,0.00005197173,0.00004010464,0.00002840855,0.00001269102,0.00002323894,0.00002456486,0.9451492,0.001575237,0.05217534,0.0008916525,0.000006439762],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0203904,0.0005518978,0.961776,0.001328694,0.0003559855,0.0001141009,0.0001441536,0.00219267,0.01314625],"genre_scores_gemma":[0.6445796,0.0002448797,0.3454857,0.001002482,0.000167255,0.0001735545,0.0003500281,0.0008782051,0.007118116],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01114715,"threshold_uncertainty_score":0.03729093,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01736942396144166,"score_gpt":0.2747101416153636,"score_spread":0.2573407176539219,"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."}}