Federated medical data - how much can deep learning models benefit? [Poster]

  • Many machine learning algorithms, like supervised Deep Learning, assume that Training Data are available in a single database. Federated Learning trains a model at each client locally, aggregates and share only the model, not the (patient-) data. Using the TensorFlow Federated Framework and data from the MIT-BIH Electrocardiogram database, we simulate two scenarios of an arrhythmia classifier (hospital and smartwatches as clients in a federated learning domain). The model quality is measured via the F1 score on a validation data set. We define the metrics Privacy Costs and Federated Benefit to evaluate the benefit of Federated Medical Data for the Deep Learning Models.

Download full text files

Export metadata

Statistics

Number of document requests

Additional Services

Share in Twitter Search Google Scholar
Metadaten
Author:Fabian StielerGND, Fabian RabeGND, Bernhard BauerORCiDGND
URN:urn:nbn:de:bvb:384-opus4-765105
Frontdoor URLhttps://opus.bibliothek.uni-augsburg.de/opus4/76510
Parent Title (English):AMIA 2020 Virtual Clinical Informatics Converence, May 19-21
Type:Conference Proceeding
Language:English
Year of first Publication:2020
Publishing Institution:Universität Augsburg
Contributing Corporation:American Medical Informatics Association
Release Date:2020/05/26
GND-Keyword:Machine Learning; Deep Learning; Federated Learning
Institutes:Fakultät für Angewandte Informatik
Fakultät für Angewandte Informatik / Institut für Informatik
Fakultät für Angewandte Informatik / Institut für Software & Systems Engineering
Fakultät für Angewandte Informatik / Institut für Informatik / Lehrstuhl für Softwaretechnik
Fakultät für Angewandte Informatik / Institut für Informatik / Lehrstuhl für Softwaretechnik / Professur Softwaremethodik für verteilte Systeme
Dewey Decimal Classification:0 Informatik, Informationswissenschaft, allgemeine Werke / 00 Informatik, Wissen, Systeme / 000 Informatik, Informationswissenschaft, allgemeine Werke
Licence (German):Deutsches Urheberrecht