{"id":"W3118325553","doi":"10.48550/arxiv.2301.12006","title":"Improved knowledge distillation by utilizing backward pass knowledge in neural networks","year":2023,"lang":"en","type":"article","venue":"arXiv (Cornell University)","topic":"Advanced Neural Network Applications","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University; University of Waterloo","funders":"","keywords":"Distillation; Divergence (linguistics); Artificial neural network; Computer science; Matching (statistics); Set (abstract data type); Domain knowledge; Artificial intelligence; Training set; Machine learning; Natural language processing; Mathematics; Statistics; Chromatography","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0001656703,0.000203762,0.0001885084,0.0002053714,0.0002077425,0.00005401477,0.0008676096,0.0001048829,0.000005733219],"category_scores_gemma":[0.00001982822,0.0002395286,0.00008031941,0.003661223,0.00007671973,0.0006470304,0.0005296282,0.0002629356,0.0001346941],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001619295,"about_ca_system_score_gemma":0.00003080212,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001621088,"about_ca_topic_score_gemma":0.0001396465,"domain_scores_codex":[0.9983717,0.0001130139,0.0002026197,0.0007679189,0.00003413554,0.0005106259],"domain_scores_gemma":[0.998759,0.0002901982,0.000102205,0.0006226824,0.00007476027,0.0001511153],"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":[0.00004738233,0.0002520299,0.01153195,0.00003500432,0.00003136051,0.00007469094,0.0005371016,0.7472711,0.001363508,0.1788848,0.007030334,0.0529408],"study_design_scores_gemma":[0.000405234,0.0000265567,0.005346619,0.00001174443,0.000006155152,0.000001562934,0.00005264712,0.9891805,0.00003813463,0.002008388,0.002676038,0.0002464251],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1973838,0.000209373,0.7984481,0.0001702568,0.0003933274,0.0003618408,0.000005033884,0.0007410568,0.002287188],"genre_scores_gemma":[0.9981694,0.0001109597,0.0002768467,0.00003190314,0.00007279784,0.00000390146,0.00002263088,0.00001919224,0.001292334],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8007857,"threshold_uncertainty_score":0.9767694,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05746430418862464,"score_gpt":0.2194172936300952,"score_spread":0.1619529894414705,"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."}}