{"id":"W3002388977","doi":"10.24908/pceea.vi0.13735","title":"AUTOMATED STUDENT ASSESSMENT: LABEL RECOGNITION IN STEM FIGURES","year":2019,"lang":"en","type":"article","venue":"Proceedings of the Canadian Engineering Education Association (CEEA)","topic":"Teaching and Learning Programming","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"MacEwan University","funders":"","keywords":"Categorization; Object (grammar); Computer science; Dimension (graph theory); Identifier; Variable (mathematics); Point (geometry); Probabilistic logic; Artificial intelligence; Cognitive neuroscience of visual object recognition; Pattern recognition (psychology); Test (biology); Series (stratigraphy); Machine learning; Data mining; Mathematics","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0008066451,0.0001052475,0.0001266958,0.0003253279,0.00009249761,0.0002419102,0.0004962352,0.0001014303,0.000006665692],"category_scores_gemma":[0.0002059452,0.0001062155,0.00004046809,0.0006255062,0.000004093681,0.0003716029,0.00004041569,0.0003100313,0.00002023471],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001896254,"about_ca_system_score_gemma":0.0004794335,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004002213,"about_ca_topic_score_gemma":0.001438512,"domain_scores_codex":[0.9989582,0.00001762516,0.0002323735,0.000206976,0.000320704,0.0002641732],"domain_scores_gemma":[0.9992107,0.00005419104,0.0002777993,0.0001103153,0.0002510622,0.00009591882],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000001464638,0.0002773709,0.8887873,0.0003969555,0.0001002999,2.078386e-7,0.009116183,0.00321329,0.001985736,0.02014915,0.008641591,0.0673305],"study_design_scores_gemma":[0.0004259587,0.00003868667,0.9110155,0.0004406972,0.0000145848,0.000002854368,0.00054743,0.06094772,0.0003150455,0.0001006552,0.02583149,0.0003193532],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9918194,0.00005425475,0.00009926504,0.003081147,0.001612593,0.000422692,0.000003420997,0.0003913289,0.002515926],"genre_scores_gemma":[0.9907278,0.000002520202,0.007574636,0.000130025,0.0000414388,0.00003696158,0.000004339612,0.00001414134,0.001468139],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.06701116,"threshold_uncertainty_score":0.6050176,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.008978495591816371,"score_gpt":0.2477271005124925,"score_spread":0.2387486049206761,"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."}}