{"id":"W3184918446","doi":"10.18653/v1/2021.repl4nlp-1.17","title":"In-Batch Negatives for Knowledge Distillation with Tightly-Coupled Teachers for Dense Retrieval","year":2021,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":119,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada; Canada First Research Excellence Fund","keywords":"Encoder; Computer science; Distillation; Ranking (information retrieval); Artificial intelligence; Knowledge transfer; Learning to rank; Information retrieval; Machine learning; Knowledge management","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00145592,0.001755269,0.00117286,0.0005612797,0.0007700715,0.00147955,0.00315736,0.001640862,0.007955279],"category_scores_gemma":[0.006699448,0.0008659789,0.0008093814,0.0006494004,0.001324986,0.00395927,0.003112664,0.003801862,0.004338209],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001182198,"about_ca_system_score_gemma":0.001825006,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005432407,"about_ca_topic_score_gemma":0.0145753,"domain_scores_codex":[0.9991902,0.0002546659,0.00004069081,0.0002354601,0.0001745026,0.0001045423],"domain_scores_gemma":[0.9983001,0.0008150215,0.0001142256,0.0004471562,0.0002208306,0.0001026019],"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":[0.0007830102,0.0004461779,0.001633286,0.0004550821,0.0001297158,0.0003054234,0.0004510797,0.3285509,0.02330733,0.04765933,0.0260308,0.5702479],"study_design_scores_gemma":[0.00003779149,0.00007995897,0.0001072087,0.0000183454,0.0000197474,0.0000608221,0.00002750166,0.9709765,0.007821436,0.01736429,0.003467299,0.0000191318],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.0148318,0.0003177983,0.9740174,0.0004083238,0.000100266,0.00008857016,0.0003173756,0.007362775,0.002555762],"genre_scores_gemma":[0.4641142,0.0002783151,0.5122764,0.0008772151,0.0002258155,0.0004576674,0.002310942,0.001800214,0.01765919],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.007955279,"threshold_uncertainty_score":0.02661306,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02914580604038951,"score_gpt":0.2862546506800759,"score_spread":0.2571088446396864,"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."}}