{"id":"W4386566901","doi":"10.18653/v1/2023.eacl-main.19","title":"Understanding Transformer Memorization Recall Through Idioms","year":2023,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":15,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"European Commission; Canadian Institute for Advanced Research","keywords":"Memorization; Computer science; Transformer; Recall; Phrase; Artificial intelligence; Natural language processing; Security token; Machine learning; Speech recognition; Cognitive psychology; 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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003522099,0.0005458416,0.0004630756,0.0007981212,0.0003769022,0.002584094,0.001165225,0.0008010994,0.002805076],"category_scores_gemma":[0.03475459,0.0005213706,0.0006430718,0.0004958341,0.0012882,0.007714496,0.001835136,0.002078017,0.0005063952],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009156486,"about_ca_system_score_gemma":0.0005238056,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001564945,"about_ca_topic_score_gemma":0.002036208,"domain_scores_codex":[0.9987508,0.000495092,0.00008361587,0.0003953876,0.0001726307,0.000102497],"domain_scores_gemma":[0.9858279,0.009784844,0.00120729,0.00221383,0.0007637873,0.0002024128],"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":[0.001454371,0.0004219456,0.2087322,0.0009906277,0.0005791317,0.001153236,0.02242088,0.1109559,0.07846139,0.1222715,0.003677229,0.4488816],"study_design_scores_gemma":[0.0001082317,0.0006164634,0.05042077,0.0001938559,0.0003295528,0.001113418,0.003815828,0.6627117,0.05686944,0.2164691,0.007188105,0.0001636226],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7285159,0.0003072856,0.263542,0.000601718,0.00002800275,0.00008695019,0.0003101687,0.0008057818,0.005802376],"genre_scores_gemma":[0.9672222,0.0001120041,0.03128187,0.00007750042,0.000009637383,0.00004618952,0.0003132664,0.0001085898,0.0008287262],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.003522099,"threshold_uncertainty_score":0.01862687,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.237016441629294,"score_gpt":0.2989166739255986,"score_spread":0.06190023229630456,"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."}}