{"id":"W2984256198","doi":"10.18653/v1/d19-6115","title":"Unlearn Dataset Bias in Natural Language Inference by Fitting the Residual","year":2019,"lang":"en","type":"article","venue":"","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":165,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"York University","keywords":"Debiasing; Residual; Computer science; Artificial intelligence; Inference; Machine learning; Benchmark (surveying); Natural language processing; Baseline (sea); Algorithm; Psychology","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0005675248,0.000120625,0.0001145526,0.00007265242,0.00005446774,0.0002562547,0.001629203,0.00005592454,0.00004899421],"category_scores_gemma":[0.0002966749,0.00007205923,0.0000179599,0.0004223328,0.00003274329,0.000644026,0.0006484601,0.0004350889,0.0001001827],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002853082,"about_ca_system_score_gemma":0.0000419,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005737777,"about_ca_topic_score_gemma":0.0001727463,"domain_scores_codex":[0.9988604,0.0001016095,0.0001787235,0.0003293708,0.0002615495,0.0002683504],"domain_scores_gemma":[0.9988256,0.0003577259,0.00007404586,0.0006884495,0.00002647187,0.0000276806],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.00004106587,0.0001855661,0.02963427,0.0001416357,0.00003830046,0.0002027923,0.0115242,0.00003069448,0.0783565,0.1230737,0.3579155,0.3988558],"study_design_scores_gemma":[0.00394795,0.000686996,0.01125813,0.001105597,0.00002960639,0.0002628687,0.004591703,0.248537,0.5897927,0.0389083,0.09613362,0.004745475],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8512947,0.0286162,0.08121464,0.02401698,0.001284184,0.00195272,0.0002836682,0.00386848,0.007468433],"genre_scores_gemma":[0.9353303,0.00000591851,0.06186121,0.00157717,0.00002342676,0.000005619341,0.0001093131,0.000006644835,0.001080422],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5114362,"threshold_uncertainty_score":0.3027492,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01741412908958099,"score_gpt":0.3055379553608717,"score_spread":0.2881238262712907,"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."}}