{"id":"W3037447448","doi":"10.3389/fams.2022.818799","title":"On Lyapunov Exponents for RNNs: Understanding Information Propagation Using Dynamical Systems Tools","year":2022,"lang":"en","type":"article","venue":"Frontiers in Applied Mathematics and Statistics","topic":"Neural dynamics and brain function","field":"Neuroscience","cited_by":23,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal; Mila - Quebec Artificial Intelligence Institute","funders":"Natural Sciences and Engineering Research Council of Canada; Institut de Valorisation des Données; Fonds de recherche du Québec – Nature et technologies; University of Washington; Canadian Institute for Advanced Research; Netherlands eScience Center; Washington Research Foundation; National Science Foundation","keywords":"Recurrent neural network; Computer science; Dynamical systems theory; Attractor; Lyapunov exponent; Lyapunov function; Initialization; Hyperparameter; Leverage (statistics); Artificial intelligence; Mathematics; Artificial neural network; Chaotic; Nonlinear system","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":[0.0002906201,0.0001302893,0.0001983221,0.0001891842,0.0003597768,0.0001605368,0.00009235722,0.00003874318,0.000003581222],"category_scores_gemma":[0.000195275,0.0001294159,0.00001767183,0.0001684204,0.00004105043,0.0001245079,0.00005760536,0.000146407,7.773871e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003959326,"about_ca_system_score_gemma":0.00002511012,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000001942013,"about_ca_topic_score_gemma":4.769241e-7,"domain_scores_codex":[0.9988892,0.00002384666,0.0003754726,0.0001902913,0.0003182201,0.0002029811],"domain_scores_gemma":[0.9992949,0.0003215066,0.0002171326,0.0001132317,0.00001362695,0.00003962773],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001002237,0.00008550866,0.000008607198,0.0002872335,0.000005156721,0.000002084376,0.0004464768,0.01509362,0.004650346,0.9743502,0.0009077681,0.00406275],"study_design_scores_gemma":[0.0005041087,0.00008931797,0.000003659015,0.00001954898,0.00001059332,0.000006365209,0.001592348,0.8422707,0.00009140973,0.1551318,0.0001507019,0.0001294402],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01890526,0.000004277028,0.9782236,0.00001839271,0.0007932793,0.00108542,0.0005409625,0.00002662705,0.0004022217],"genre_scores_gemma":[0.7434947,0.00001435677,0.2558244,0.0001699672,0.00002249143,0.0002612779,0.0001512391,0.00002752549,0.00003397526],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.827177,"threshold_uncertainty_score":0.5277429,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06218042407756191,"score_gpt":0.2571484869486104,"score_spread":0.1949680628710485,"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."}}