{"id":"W4399601886","doi":"10.1145/3672457","title":"<i>GIST</i> : Generated Inputs Sets Transferability in Deep Learning","year":2024,"lang":"en","type":"article","venue":"ACM Transactions on Software Engineering and Methodology","topic":"Adversarial Robustness in Machine Learning","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Polytechnique Montréal","funders":"","keywords":"Computer science; Scratch; Test set; GiST; Testability; Test (biology); Set (abstract data type); Artificial intelligence; Property (philosophy); Artificial neural network; Machine learning; Transfer of learning; Data mining; Reliability engineering; Programming language","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":[0.005787595,0.001395239,0.0007519177,0.00128162,0.000418994,0.001550769,0.002919201,0.001625041,0.003737976],"category_scores_gemma":[0.03102725,0.0005961552,0.001146479,0.0008188823,0.003451832,0.003668059,0.003595756,0.003477153,0.0007125886],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001588322,"about_ca_system_score_gemma":0.001134374,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001566798,"about_ca_topic_score_gemma":0.001241733,"domain_scores_codex":[0.9949999,0.00194466,0.0003365253,0.0008821764,0.00153961,0.0002972425],"domain_scores_gemma":[0.9816625,0.01046956,0.001319248,0.005122036,0.00115137,0.0002753261],"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":[0.0005541619,0.0002600284,0.00462751,0.0004464395,0.0001730675,0.0006205254,0.0004141628,0.478938,0.0275212,0.07873897,0.007154218,0.4005517],"study_design_scores_gemma":[0.00003177239,0.0002005998,0.0004555425,0.00006191641,0.0000244566,0.0001657635,0.0000235881,0.9196472,0.02626254,0.05075969,0.002344941,0.0000219604],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01853616,0.0002891657,0.9744059,0.0004841893,0.00005188402,0.0001378044,0.0001267783,0.003771059,0.002197132],"genre_scores_gemma":[0.6581245,0.0003458057,0.3366622,0.000649351,0.00009589452,0.0004351142,0.00068726,0.0008469824,0.002152981],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.005787595,"threshold_uncertainty_score":0.03060812,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04779516566455851,"score_gpt":0.3112324647310983,"score_spread":0.2634372990665398,"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."}}