{"id":"W3171596153","doi":"10.1109/jsac.2021.3087272","title":"LOSP: Overlap Synchronization Parallel With Local Compensation for Fast Distributed Training","year":2021,"lang":"en","type":"article","venue":"IEEE Journal on Selected Areas in Communications","topic":"Stochastic Gradient Optimization Techniques","field":"Computer Science","cited_by":21,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"National Key Research and Development Program of China; Research Grants Council, University Grants Committee; China Postdoctoral Science Foundation; Impact Fund; Science, Technology and Innovation Commission of Shenzhen Municipality; National Natural Science Foundation of China","keywords":"Computer science; Scalability; Synchronization (alternating current); Computation; Speedup; Distributed computing; Overhead (engineering); Convergence (economics); Stochastic gradient descent; Rate of convergence; Compensation (psychology); Data synchronization; Mathematical optimization; Parallel computing; Algorithm; Key (lock); Computer network; Artificial intelligence; 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":[0.001122499,0.0009177732,0.001135864,0.0003555401,0.0005884733,0.0005875189,0.001815226,0.000765622,0.0024227],"category_scores_gemma":[0.002440562,0.0004259979,0.0004561743,0.0005839373,0.0006978395,0.001334504,0.002022063,0.00132865,0.0008170732],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004609751,"about_ca_system_score_gemma":0.002208167,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003956833,"about_ca_topic_score_gemma":0.005111638,"domain_scores_codex":[0.9993016,0.0001603212,0.00004309014,0.0001817567,0.000217668,0.00009562547],"domain_scores_gemma":[0.9991668,0.0002910636,0.00008847619,0.000222126,0.0001432753,0.00008814083],"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":[0.0007896101,0.0003208776,0.002073382,0.0002218239,0.00009843415,0.0002355515,0.0002829713,0.5344444,0.01943559,0.01228613,0.01097003,0.4188412],"study_design_scores_gemma":[0.00005733607,0.0000759518,0.000168674,0.000004554687,0.000007845612,0.00003037879,0.0000155264,0.9941133,0.002391105,0.001885017,0.001243509,0.000006783695],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01516369,0.0001783649,0.9809994,0.00015159,0.00007492608,0.00005253585,0.00003211939,0.002210141,0.001137267],"genre_scores_gemma":[0.600266,0.0002264024,0.3926405,0.0003085647,0.0001543953,0.000357829,0.000363808,0.0004627568,0.005219696],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.003956833,"threshold_uncertainty_score":0.008104682,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03917666790837083,"score_gpt":0.2856775653390298,"score_spread":0.246500897430659,"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."}}