{"id":"W4385572282","doi":"10.18653/v1/2023.findings-acl.496","title":"Impact of Adversarial Training on Robustness and Generalizability of Language Models","year":2023,"lang":"en","type":"article","venue":"","topic":"Adversarial Robustness in Machine Learning","field":"Computer Science","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University","funders":"","keywords":"Adversarial system; Generalizability theory; Computer science; Embedding; Robustness (evolution); Artificial intelligence; Generalization; Transformer; Machine learning; Training set; Mathematics; Engineering","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0006499648,0.0001314845,0.0002926327,0.0001767341,0.00004845027,0.00001869931,0.0004344529,0.00009401918,0.00002534641],"category_scores_gemma":[0.0001708062,0.0001064667,0.0001092771,0.0005394465,0.00008273972,0.0003420513,0.000289003,0.0001690964,0.000001465156],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003097714,"about_ca_system_score_gemma":0.00008711948,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004069539,"about_ca_topic_score_gemma":0.000007060371,"domain_scores_codex":[0.9987856,0.0001322462,0.0002686342,0.0003115965,0.0002739993,0.0002278987],"domain_scores_gemma":[0.9990506,0.000251336,0.0001306789,0.0004440039,0.00005749926,0.00006584566],"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":[0.00002608336,0.00002429208,0.001206336,0.0000194592,0.00002779623,0.000005204291,0.004493932,0.960512,0.001460398,0.02305307,0.00004650051,0.009124935],"study_design_scores_gemma":[0.0005299731,0.0001311652,0.009676239,0.00001500921,0.000005652766,0.000002299844,0.0002476487,0.9876339,0.0004268196,0.001220146,0.0000013782,0.0001097865],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5842615,0.000006264507,0.4144451,0.00007268513,0.00009496195,0.00007010471,0.00000335957,0.0001016832,0.000944323],"genre_scores_gemma":[0.9750231,0.000003853243,0.02482382,0.00001391322,0.00004992867,0.000002535279,0.000003531883,0.000009725341,0.00006957283],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3907616,"threshold_uncertainty_score":0.4341587,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04219023531837281,"score_gpt":0.3193435283298187,"score_spread":0.2771532930114459,"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."}}