{"id":"W4387735129","doi":"10.29007/6nqs","title":"ARCH-COMP23 Category Report: Falsification","year":2023,"lang":"en","type":"article","venue":"EPiC series in computing","topic":"Physical Unclonable Functions (PUFs) and Hardware Security","field":"Computer Science","cited_by":15,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University","funders":"JST-Mirai Program; Core Research for Evolutional Science and Technology; ACT-X; Japan Society for the Promotion of Science; European Commission; Natural Sciences and Engineering Research Council of Canada; Defense Advanced Research Projects Agency; Electronic Components and Systems for European Leadership; Exploratory Research for Advanced Technology; Division of Civil, Mechanical and Manufacturing Innovation; National Science Foundation","keywords":"Benchmark (surveying); Arch; Computer science; Competition (biology); Temporal logic; Cyber-physical system; Software engineering; Data science; Artificial intelligence; Theoretical computer science; Engineering; Civil engineering","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"}],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0196413,0.002953922,0.00160065,0.00440285,0.003423885,0.009612543,0.005081551,0.005039667,0.07601599],"category_scores_gemma":[0.04506953,0.0009388309,0.0026603,0.003190147,0.002161759,0.006853102,0.006787135,0.005347886,0.05132535],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00642676,"about_ca_system_score_gemma":0.007739023,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0125451,"about_ca_topic_score_gemma":0.01109209,"domain_scores_codex":[0.9733925,0.006268633,0.001077169,0.002644708,0.01442429,0.002192717],"domain_scores_gemma":[0.9162331,0.02343143,0.002475038,0.0206091,0.03300473,0.004246633],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.002154646,0.0007161994,0.002770971,0.001011866,0.000133583,0.0003387463,0.0001795735,0.01273486,0.003241091,0.03279,0.8918507,0.05207789],"study_design_scores_gemma":[0.0008925732,0.00133732,0.003124783,0.0005007471,0.0001460546,0.000960791,0.0007356043,0.05644134,0.03126137,0.05975964,0.844592,0.0002477971],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.07165765,0.00586366,0.1825929,0.02421292,0.02384862,0.002762308,0.2298142,0.1048667,0.3543811],"genre_scores_gemma":[0.2439145,0.002135589,0.09673135,0.006882177,0.002607509,0.001990761,0.4821198,0.0231899,0.1404285],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.07601599,"threshold_uncertainty_score":0.2542988,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02443590830536669,"score_gpt":0.2745188595835987,"score_spread":0.250082951278232,"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."}}