(A) experimental survey of federated learning on the medical domain

Abstract

Federated learning (FL) is an active area of research. One of the most suitable areas for adopting FLis the medical domain, where patient privacy must be respected. Previous research, however, does notfully consider who will most likely use FL in the medical domain. It is not the hospitals who are eagerto adopt FL, but the service providers such as IT companies who want to develop artificial intelligencemodels with real patient records. Moreover, service providers would prefer to focus on maximizing theperformance of the models at the lowest cost possible. In this work, we propose empirical benchmarks ofFL methods considering both performance and monetary cost with three real-world datasets: electronichealth records, skin cancer images, and electrocardiogram datasets. We also propose Federated learningwith Proximal regularization eXcept local Normalization (FedPxN), which, using a simple combinationof FedProx and FedBN, outperforms all other FL algorithms while consuming only slightly more powerthan the most power efficient method.