{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0001078776,0.00009521791,0.0001114831,0.00002255643,0.0001186165,0.000250649,0.000635769,0.0000362999,0.00001599931],"category_scores_gemma":[0.0001970871,0.000089324,0.00004444044,0.0001416259,0.000008035755,0.000485546,0.0002566557,0.00009186014,0.00002749659],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001799725,"about_ca_system_score_gemma":0.00004625859,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004569367,"about_ca_topic_score_gemma":0.00002051433,"domain_scores_codex":[0.9990374,0.00001431546,0.0002079626,0.0003602021,0.0001320577,0.0002480102],"domain_scores_gemma":[0.9993942,0.0001653891,0.0000517532,0.0002309431,0.00006005041,0.00009765562],"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":[0.000007312496,0.00002777528,0.004984529,0.0001055974,0.0000275038,0.000009798353,0.01109455,0.004333694,0.01047697,0.3987543,0.0002104179,0.5699676],"study_design_scores_gemma":[0.0001563188,0.00004045817,0.0004213571,0.00001066958,0.000001561135,0.000001069843,0.00002753992,0.995522,0.0008165872,0.00255396,0.0003244987,0.0001239888],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.08990247,0.00001601327,0.903221,0.003388924,0.00008943948,0.0001498811,5.069712e-7,0.000249056,0.002982783],"genre_scores_gemma":[0.6414949,5.736424e-7,0.3570764,0.001241174,0.00005083238,0.00001348326,5.635258e-7,0.000005122051,0.0001169284],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9911883,"threshold_uncertainty_score":0.3642527,"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."}}