{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0007229895,0.0001494688,0.0002160285,0.0002317383,0.0002718256,0.0001245267,0.0006591649,0.00005888001,0.00001008239],"category_scores_gemma":[0.0002396064,0.0001562509,0.00007015654,0.001665537,0.00007890379,0.0005052234,0.0005194773,0.00026648,0.0001834925],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007815353,"about_ca_system_score_gemma":0.00008971426,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000161377,"about_ca_topic_score_gemma":0.00004175748,"domain_scores_codex":[0.9982181,0.0001074729,0.0004161798,0.0005449065,0.0002821359,0.0004311911],"domain_scores_gemma":[0.9987103,0.0002646872,0.0001334429,0.0007329353,0.0000821133,0.00007648501],"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":[0.00002177983,0.0003777842,0.01532503,0.0001455772,0.00004382402,0.001173391,0.008516927,0.009479051,0.001570305,0.7733496,0.00939415,0.1806026],"study_design_scores_gemma":[0.0008657739,0.0001956963,0.1872735,0.0001490493,0.00000981173,0.0003894311,0.0007489095,0.4630533,0.002622003,0.2529877,0.09069411,0.001010674],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5327868,0.0001350946,0.4307043,0.004845433,0.004313786,0.0004268118,0.000003318689,0.002463926,0.02432059],"genre_scores_gemma":[0.9932486,0.00001144984,0.005672498,0.0001417596,0.0003086895,0.00001269681,0.00002728017,0.00001158624,0.0005654016],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5203618,"threshold_uncertainty_score":0.6371726,"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."}}