{"id":"W3035038672","doi":"10.18653/v1/2020.acl-main.204","title":"DeeBERT: Dynamic Early Exiting for Accelerating BERT Inference","year":2020,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":301,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo; Vector Institute","funders":"Vector Institute; Natural Sciences and Engineering Research Council of Canada; Government of Canada; Canadian Institute for Advanced Research","keywords":"Inference; Computer science; Transformer; Language model; Redundancy (engineering); Artificial intelligence; Machine learning; Operating system; Engineering","routes":{"ca_aff":true,"ca_fund":true,"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":[0.001873976,0.001999327,0.00124931,0.001392576,0.0008672692,0.001821755,0.003288496,0.001838304,0.01076242],"category_scores_gemma":[0.01125251,0.001242795,0.001288172,0.0009307718,0.0008216222,0.005520203,0.003142605,0.004623414,0.005447936],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001262262,"about_ca_system_score_gemma":0.002217198,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01194943,"about_ca_topic_score_gemma":0.02968272,"domain_scores_codex":[0.9990557,0.0002157725,0.00005726882,0.0002950476,0.0002426563,0.0001336565],"domain_scores_gemma":[0.9975017,0.001370215,0.0001181243,0.0005075997,0.0003532481,0.0001491134],"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":[0.001119746,0.0002759195,0.004740184,0.0004085226,0.0002362869,0.0003643305,0.0005149883,0.1959019,0.02042255,0.03581836,0.04730601,0.6928912],"study_design_scores_gemma":[0.00003995478,0.00003569089,0.0001953178,0.00001552397,0.00002196136,0.00005568617,0.0000319895,0.9761317,0.005982973,0.01294112,0.0045294,0.00001864353],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.009571702,0.0004315911,0.9644204,0.0003155717,0.0001688183,0.0001003873,0.000520601,0.02273075,0.001740136],"genre_scores_gemma":[0.3148634,0.0005258096,0.6637363,0.0007053777,0.0002250567,0.0003175901,0.004135612,0.004512577,0.01097833],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01194943,"threshold_uncertainty_score":0.03600389,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07716608598601515,"score_gpt":0.2962549808226925,"score_spread":0.2190888948366773,"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."}}