{"id":"W3204470331","doi":"10.18653/v1/2022.spnlp-1.7","title":"Predicting Attention Sparsity in Transformers","year":2022,"lang":"en","type":"preprint","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":11,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal","funders":"","keywords":"Computer science; Quadratic equation; Softmax function; Transformer; Machine translation; Computation; Encoder; Cluster analysis; Quantization (signal processing); Algorithm; Artificial intelligence; Language model; Theoretical computer science; Machine learning; Mathematics; Artificial neural network","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":[0.0004836645,0.0001133187,0.0001478443,0.00015165,0.00006341996,0.00007357873,0.0008247051,0.00008803786,0.0001074591],"category_scores_gemma":[0.000009021791,0.0001249567,0.00008433768,0.0001381556,0.000007160745,0.0001672141,0.001232834,0.0006153085,0.000004827299],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001666157,"about_ca_system_score_gemma":0.00008547313,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007414995,"about_ca_topic_score_gemma":0.0002154423,"domain_scores_codex":[0.9986736,0.00006109084,0.0002524127,0.0005249112,0.0002887096,0.0001993274],"domain_scores_gemma":[0.9994532,0.00002133182,0.00006253277,0.0004124578,0.00001400076,0.00003649439],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0000157671,0.0003192351,0.3654943,0.0006398622,0.00009400858,0.0001366682,0.01085633,0.1919599,0.0002771038,0.09771597,0.0002380871,0.3322527],"study_design_scores_gemma":[0.000164885,0.0000112749,0.02236287,0.00003856784,0.000003855216,0.000002733914,0.0001288223,0.9713826,0.00002244454,0.005392363,0.0003016062,0.000187938],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3238742,0.00002891204,0.6575594,0.0005977696,0.0008144097,0.0001927465,0.000001554789,0.0001699232,0.01676113],"genre_scores_gemma":[0.9641171,0.00001686086,0.03519711,0.00009618443,0.00004063194,0.00003117104,0.000008150089,0.000005536678,0.0004872596],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7794227,"threshold_uncertainty_score":0.5095588,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03754596689072241,"score_gpt":0.2597071815228978,"score_spread":0.2221612146321754,"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."}}