{"id":"W6966943977","doi":"10.48448/fpbg-0f77","title":"RoAST: Robustifying Language Models via Adversarial Perturbation with Selective Training","year":2023,"lang":"en","type":"other","venue":"Open MIND","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Kootenay Association for Science & Technology","funders":"","keywords":"Robustness (evolution); Adversarial system; Perturbation (astronomy); Training set; Language model","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":[0.001391908,0.001647478,0.0008380153,0.0004569675,0.0004465794,0.0008581876,0.001608676,0.001321465,0.003580582],"category_scores_gemma":[0.007252176,0.0003977989,0.0006982576,0.0003333271,0.00119156,0.001652629,0.002130634,0.002461513,0.001730865],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000510848,"about_ca_system_score_gemma":0.0009031348,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003220212,"about_ca_topic_score_gemma":0.004148914,"domain_scores_codex":[0.9991189,0.0002854717,0.00004184885,0.0002639459,0.0002108432,0.00007911687],"domain_scores_gemma":[0.998111,0.001111047,0.0001451363,0.0003787381,0.0001940685,0.00006003753],"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":[0.0001837059,0.00007860274,0.0008076721,0.0001893308,0.0001212289,0.000227524,0.0001719734,0.7901983,0.02096032,0.01458726,0.006491923,0.1659821],"study_design_scores_gemma":[0.00001189854,0.00003684617,0.00009040015,0.00001144078,0.00001097716,0.00004169277,0.00001335607,0.9873732,0.005428551,0.005160321,0.001808648,0.00001263622],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01260788,0.0004126916,0.9788933,0.0002710735,0.0001150542,0.00006481705,0.0001259167,0.00395568,0.003553545],"genre_scores_gemma":[0.6325296,0.0005364367,0.353368,0.0008046185,0.0001522505,0.0002711962,0.0007311042,0.001514809,0.01009201],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.003580582,"threshold_uncertainty_score":0.01197821,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07054775804171788,"score_gpt":0.3084281468722341,"score_spread":0.2378803888305163,"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."}}