{"id":"W3175052694","doi":"10.18653/v1/2021.acl-long.377","title":"Turn the Combination Lock: Learnable Textual Backdoor Attacks via Word Substitution","year":2021,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":93,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"National Key Research and Development Program of China","keywords":"Backdoor; Substitution (logic); Computer science; Lock (firearm); Word (group theory); Computational linguistics; Natural language processing; Artificial intelligence; Linguistics; History; Programming language; Philosophy; Computer security","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":[0.001912165,0.001218571,0.001785289,0.001107525,0.001104886,0.002245805,0.002179952,0.001898013,0.009940633],"category_scores_gemma":[0.007619973,0.0007792272,0.001140633,0.001177721,0.00207083,0.008061816,0.006527551,0.003024586,0.004258008],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007702814,"about_ca_system_score_gemma":0.001299681,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005727278,"about_ca_topic_score_gemma":0.00111032,"domain_scores_codex":[0.9964318,0.001041866,0.000260526,0.0006888222,0.0008964952,0.0006805167],"domain_scores_gemma":[0.9935035,0.002776259,0.0004105761,0.002687625,0.0004056244,0.0002163592],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.003931299,0.0009051402,0.003289352,0.0008402748,0.0005607634,0.0009558066,0.0007778816,0.06480153,0.05118665,0.1713831,0.07930561,0.6220627],"study_design_scores_gemma":[0.0004176394,0.0006039884,0.0004795173,0.0001051007,0.0002462684,0.000587126,0.0001868759,0.6087631,0.04153801,0.3261041,0.02086099,0.0001073042],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.08741924,0.002811057,0.8647347,0.002344814,0.001060774,0.0003613859,0.0009948389,0.02279363,0.01747952],"genre_scores_gemma":[0.882551,0.0005186163,0.1026487,0.001121059,0.0003706976,0.0002839174,0.0009295978,0.001154222,0.01042232],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.009940633,"threshold_uncertainty_score":0.03325468,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02142620390299724,"score_gpt":0.2467543115358606,"score_spread":0.2253281076328634,"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."}}