{"id":"W4389519618","doi":"10.18653/v1/2023.emnlp-main.125","title":"Self-Influence Guided Data Reweighting for Language Model Pre-training","year":2023,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Canadian Institute for Advanced Research; Polytechnique Montréal; Mila - Quebec Artificial Intelligence Institute","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institute for Advanced Research","keywords":"Computer science; Novelty; Artificial intelligence; Language model; Context (archaeology); Task (project management); Machine learning; Relevance (law); Sample (material); Training set; Stability (learning theory); Point (geometry); Data modeling; Natural language processing","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":[0.0008433714,0.000105715,0.000123322,0.00008103794,0.0001342921,0.0001471612,0.002170481,0.0000458444,0.000002551101],"category_scores_gemma":[0.000235376,0.00009616472,0.00002965637,0.0003083106,0.000008331963,0.0008663485,0.001163397,0.00007355625,0.000031711],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000214069,"about_ca_system_score_gemma":0.0001109434,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003392064,"about_ca_topic_score_gemma":0.0000121142,"domain_scores_codex":[0.9985222,0.00001718837,0.000253379,0.0006027549,0.0002207418,0.0003837409],"domain_scores_gemma":[0.9980779,0.0001558526,0.00005879403,0.001582179,0.00005591848,0.00006931896],"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":[0.000006359528,0.00005261902,0.0003707606,0.0003070286,0.00008570648,0.00005301882,0.05367995,0.4927031,0.01139576,0.3038382,0.0205815,0.116926],"study_design_scores_gemma":[0.0001517011,0.000006633452,0.00003349303,0.00001594529,0.000003752722,0.000004387885,0.0001262528,0.9939656,0.0003705572,0.00444602,0.0007427579,0.000132916],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0516747,0.00002112952,0.9420639,0.0007432132,0.0001166173,0.0002209276,0.000008282756,0.001554714,0.003596551],"genre_scores_gemma":[0.261045,0.000003807854,0.7365611,0.0003780307,0.00009011829,0.00002218032,0.00001651536,0.0000116295,0.001871713],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.5012625,"threshold_uncertainty_score":0.403333,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1264040835937797,"score_gpt":0.3563720179076763,"score_spread":0.2299679343138966,"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."}}