{"id":"W3023528699","doi":"10.18653/v1/2021.eacl-main.39","title":"AdapterFusion: Non-Destructive Task Composition for Transfer Learning","year":2021,"lang":"en","type":"preprint","venue":"","topic":"Domain Adaptation and Few-Shot Learning","field":"Computer Science","cited_by":41,"is_retracted":false,"has_abstract":true,"ca_institutions":"Canadian Institute for Advanced Research","funders":"Deutsche Forschungsgemeinschaft; Samsung; DeepMind; Samsung Advanced Institute of Technology","keywords":"Computer science; Forgetting; Multi-task learning; Exploit; Classifier (UML); Task (project management); Artificial intelligence; Machine learning; Source code; Transfer of learning","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":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0003311145,0.0003053471,0.0003793587,0.0001525042,0.0003100838,0.0006744983,0.0006770316,0.0002662452,0.0001028078],"category_scores_gemma":[0.00002597829,0.0003139321,0.0002844402,0.0001867559,0.00004119972,0.0003449603,0.0005169741,0.0008015381,0.00002331123],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008495619,"about_ca_system_score_gemma":0.0002144846,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003774763,"about_ca_topic_score_gemma":0.000009732619,"domain_scores_codex":[0.9979217,0.0001509857,0.0003736693,0.0008819851,0.0003502848,0.0003213747],"domain_scores_gemma":[0.9988031,0.0001794012,0.0001028364,0.0004746437,0.0003093432,0.0001306268],"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":[0.0002127124,0.0004367174,0.0003214146,0.0008693866,0.0005420481,0.0001500468,0.04028283,0.2968427,0.0686753,0.2447673,0.0008297655,0.3460698],"study_design_scores_gemma":[0.001270232,0.0002172193,0.001696923,0.0003645906,0.00005074371,0.00004344482,0.001309877,0.9799175,0.004072736,0.003613741,0.00650212,0.0009408892],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.03474248,0.00008098504,0.9557062,0.0006501408,0.0009767395,0.000528058,0.00000378272,0.0003005631,0.007011041],"genre_scores_gemma":[0.7970858,0.00003228405,0.2012052,0.0003858536,0.0001460467,0.0001055593,0.0001476244,0.00002586044,0.0008657571],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.7623433,"threshold_uncertainty_score":0.9999313,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02071584240950941,"score_gpt":0.2561504059373402,"score_spread":0.2354345635278308,"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."}}