{"id":"W3187199493","doi":"10.1109/cec45853.2021.9504998","title":"Evolving Simple Solutions to the CIFAR-10 Benchmark using Tangled Program Graphs","year":2021,"lang":"en","type":"article","venue":"","topic":"Evolutionary Algorithms and Applications","field":"Computer Science","cited_by":11,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University","funders":"","keywords":"Computer science; Benchmark (surveying); Benchmarking; Genetic programming; Class (philosophy); Simplicity; Machine learning; Modularity (biology); Artificial intelligence; Cardinality (data modeling); Graph; Contextual image classification; Theoretical computer science; Data mining; Image (mathematics)","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":[0.001262047,0.001088143,0.0005767423,0.001082586,0.000605716,0.0008347514,0.00150562,0.001447215,0.002182631],"category_scores_gemma":[0.005030812,0.0002926301,0.0006111524,0.000958468,0.0007652547,0.0009387061,0.0008665101,0.001397416,0.0002979582],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001450533,"about_ca_system_score_gemma":0.001225696,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007442502,"about_ca_topic_score_gemma":0.01339104,"domain_scores_codex":[0.9995512,0.0001451805,0.00001983727,0.0001123765,0.00009671115,0.00007464818],"domain_scores_gemma":[0.9976826,0.001602448,0.0001128265,0.0002257594,0.0002552988,0.000120943],"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":[0.0001332431,0.0001914354,0.002023196,0.0001596475,0.00006920363,0.0001580086,0.0001061831,0.9335293,0.002309152,0.008645593,0.00576565,0.04690934],"study_design_scores_gemma":[0.00004014283,0.00007884252,0.000284918,0.0000132935,0.00001202631,0.00003239053,0.00005287162,0.9887165,0.001516702,0.007797159,0.001447842,0.000007283022],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7811542,0.001068907,0.1906282,0.001652354,0.0002401914,0.0002413501,0.001271891,0.003441245,0.02030179],"genre_scores_gemma":[0.724149,0.0003177377,0.2678854,0.0003532751,0.00003243029,0.0003257103,0.002281951,0.0004432368,0.004211159],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.007442502,"threshold_uncertainty_score":0.01479834,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0301849842186758,"score_gpt":0.2877595361043339,"score_spread":0.2575745518856581,"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."}}