{"id":"W4385570529","doi":"10.18653/v1/2023.findings-acl.421","title":"Residual Prompt Tuning: improving prompt tuning with residual reparameterization","year":2023,"lang":"en","type":"article","venue":"","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":21,"is_retracted":false,"has_abstract":true,"ca_institutions":"Vector Institute; University of Toronto","funders":"","keywords":"Residual; Computer science; Initialization; Benchmark (surveying); Fine-tuning; Base (topology); Algorithm; Mathematics; Physics","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":[0.0007258581,0.0002504738,0.0002226804,0.000391452,0.0003032275,0.0006054555,0.001067042,0.0001230815,0.00001056988],"category_scores_gemma":[0.0003004911,0.000192328,0.00002931836,0.00165676,0.00007483119,0.001334988,0.0006736161,0.0003079018,0.00003060119],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006870714,"about_ca_system_score_gemma":0.0002051393,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00009771506,"about_ca_topic_score_gemma":0.00002277546,"domain_scores_codex":[0.9976429,0.00009856025,0.000330665,0.0007870118,0.000598575,0.0005422362],"domain_scores_gemma":[0.9985916,0.0001103769,0.0002027318,0.0007936187,0.0001984642,0.0001032333],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.000194809,0.0001788927,0.003268029,0.0006504536,0.0001269107,0.001100192,0.01093812,0.0002261691,0.5840504,0.1201334,0.01484431,0.2642884],"study_design_scores_gemma":[0.001525728,0.002061818,0.002614625,0.0009657133,0.00006461739,0.0004648643,0.0006397558,0.334392,0.6147586,0.03843671,0.001487917,0.002587707],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1185591,0.0001920895,0.8644506,0.002540557,0.0001824452,0.0008067408,0.000002813859,0.01142738,0.001838201],"genre_scores_gemma":[0.3650999,0.000002973476,0.6324115,0.0001780489,0.00009605082,0.00005978845,0.00002361808,0.0000343799,0.002093695],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.3341658,"threshold_uncertainty_score":0.7842907,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01721700783509059,"score_gpt":0.2636806702699052,"score_spread":0.2464636624348146,"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."}}