Federating Medical Data: How much can Deep Learning Models Benefit?

3:30 PM–4:30 PM May 19, 2020 (Conference Time: US - Pacific)

3:30 PM–4:30 PM May 19, 2020

Regency Prefunction


Abstract: Patient data sets can't always be aggregated across different institutions. TensorFlow Federated is a framework to enable Deep Learning on decentralized data. We compare the performance of a Neural Network classifying Electrocardiogram records when trained on centralized data versus two simulated federations of institutions. Based on the gathered data, we measure the model performance cost of keeping the data federated as well as the benefit, a single institution has from joining the federation.

Describe the new knowledge and additional skills the participant will gain after attending your presentation.: The attendees will be able to decide whether it is necessary to aggregate patient data in a single database for a machine learning project, or whether the technique of federated deep learning might provide a comparable machine learning model whilst leaving patient data at their respective sources.


Fabian Rabe (Presenter)
University of Augsburg

Fabian Stieler, University of Augsburg
Bernhard Bauer, University of Augsburg