{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001695863,0.002586627,0.001075633,0.000759134,0.0006386696,0.001477403,0.002415254,0.002182625,0.009473467],"category_scores_gemma":[0.0121599,0.0007305218,0.0009785574,0.0005652232,0.0009990323,0.005106525,0.002867686,0.004775132,0.005221465],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008680787,"about_ca_system_score_gemma":0.001386308,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001973066,"about_ca_topic_score_gemma":0.004519576,"domain_scores_codex":[0.9988701,0.0003161064,0.00006401784,0.0004503872,0.0001651389,0.0001341829],"domain_scores_gemma":[0.998192,0.000864006,0.0001038252,0.0004621519,0.0002480256,0.0001299543],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.001535454,0.0006602098,0.003649973,0.0006835752,0.0001615901,0.000573424,0.0007441072,0.2391605,0.04357896,0.01543157,0.03763406,0.6561866],"study_design_scores_gemma":[0.0001649525,0.0003145977,0.0004912502,0.00005721634,0.00004350818,0.0001456706,0.0001653197,0.9517955,0.01627174,0.02300855,0.007484191,0.00005749524],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.08100789,0.001276879,0.8470649,0.0006622523,0.0007105072,0.0002975387,0.0007382362,0.06189466,0.006347144],"genre_scores_gemma":[0.6984284,0.0003872957,0.2806001,0.001400723,0.0001854049,0.0005219368,0.002620292,0.005822462,0.01003338],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.009473467,"threshold_uncertainty_score":0.03169191,"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."}}