{"id":"W3213820586","doi":"","title":"How to Select One Among All? An Extensive Empirical Study Towards the Robustness of Knowledge Distillation in Natural Language Understanding.","year":2021,"lang":"en","type":"article","venue":"Empirical Methods in Natural Language Processing","topic":"Adversarial Robustness in Machine Learning","field":"Computer Science","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo; Queen's University","funders":"","keywords":"Robustness (evolution); Computer science; Adversarial system; Distillation; Artificial intelligence; Machine learning; Benchmark (surveying); Domain knowledge; Natural language; Artificial neural network; Theoretical computer science","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":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.002810686,0.0004113159,0.0007588373,0.0004516329,0.0002164434,0.0004467233,0.001141025,0.0002071784,0.00000732534],"category_scores_gemma":[0.007180945,0.0003079489,0.0001200775,0.003927593,0.000155722,0.001075802,0.0009116746,0.001693806,6.763905e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005647023,"about_ca_system_score_gemma":0.0003197724,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001137195,"about_ca_topic_score_gemma":0.001030266,"domain_scores_codex":[0.9934838,0.003352826,0.0006683785,0.001104063,0.0007057841,0.0006851104],"domain_scores_gemma":[0.9969469,0.001593173,0.0003098342,0.0006558951,0.0003338946,0.0001603305],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000234778,0.001084221,0.06356785,0.0002860995,0.0001022507,0.0006473039,0.2178968,0.0150819,0.004791423,0.0001946324,0.00003735459,0.6960754],"study_design_scores_gemma":[0.001233748,0.0001870639,0.1691274,0.0003470544,0.0000536111,0.00004367035,0.04061834,0.7854391,0.001983682,0.0002942358,0.00001911679,0.000653034],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4414481,0.004485593,0.5513006,0.001726963,0.000418104,0.0004351607,0.000001016823,0.0001062832,0.00007819655],"genre_scores_gemma":[0.7291704,0.000001595904,0.2702582,0.0002260201,0.00018634,0.00002286973,0.000008113249,0.00003063695,0.00009588261],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7703572,"threshold_uncertainty_score":0.9999372,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1018436876354653,"score_gpt":0.4440969145619636,"score_spread":0.3422532269264983,"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."}}