Trustworthy Hierarchical Federated Learning for Digital Healthcare

Healthcare institutions and medical device manufacturers are under regulatory obligations to safeguard and protect the privacy of data they acquire from patients. This limits their ability to share the data with other institutions to collectively train machine learning models. Due to its ability in preserving the privacy of data used in training models, Federated Learning (FL) has been proposed as a tool in healthcare that mitigates some of the privacy concerns. However, the presence of less-engaging clients; or free-riders; in such environments is a major concern. Such clients reap the benefits of the global model (1) without actively participating in the learning rounds, and (2) by not contributing their system and data resources. In this work, we propose a mechanism that minimizes the presence of free-riders in such environments. Experimental results shows the effectiveness of our approach.