{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001446971,0.0006411118,0.0007072726,0.0007685837,0.0003623178,0.001153679,0.001088252,0.0008537658,0.003224191],"category_scores_gemma":[0.01396362,0.0004939168,0.0005750138,0.0006988605,0.0009285516,0.002828366,0.001411328,0.001572425,0.0007930925],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001233027,"about_ca_system_score_gemma":0.000902429,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00672569,"about_ca_topic_score_gemma":0.0109382,"domain_scores_codex":[0.9994018,0.0002165163,0.00002228232,0.0001798688,0.0001028952,0.00007656629],"domain_scores_gemma":[0.9941214,0.004460748,0.000265299,0.0006193859,0.0004046462,0.0001285843],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0009064786,0.0001215758,0.00627574,0.0002003219,0.0000764558,0.0001859011,0.0005060531,0.6762547,0.01731042,0.05896426,0.006529727,0.2326684],"study_design_scores_gemma":[0.00001081704,0.0000213229,0.0003753566,0.00000429943,0.000006456055,0.00001978107,0.00001560924,0.9761486,0.002288085,0.02080145,0.000303651,0.000004508837],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1706653,0.0003070081,0.8227484,0.0006940172,0.00003740251,0.00006049418,0.000467558,0.002577133,0.002442832],"genre_scores_gemma":[0.9252241,0.0002198766,0.06957336,0.0002062184,0.0000483014,0.00008760211,0.0007265266,0.0002672821,0.003646651],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.00672569,"threshold_uncertainty_score":0.01337308,"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."}}