Pioneering the Hybridization of Federated Learning in Human Activity Recognition

The Internet of Things (IoT) nowadays greatly benefits from Artificial Intelligence (AI) algorithms implemented in the edge, because of their efficiency and the reduction of costs that they imply. The advent of Federated Learning (FL) has made possible the combination of the advantages of edge-AI, among which the privacy of users, as their data is not shared with the cloud, with the collective intelligence. However, FL is known to have worse performance compared to its centralized counterpart, which may not be tolerable in certain cases. In this paper, we propose a hybrid framework for FL, imagining a number of clients that are willing to share part of their data. We envisioned two types of Hybridization: vertical and horizontal. The goal of this paper is to assess whether a small hybridization can bring advantages to the overall performance of the whole FL procedure in terms of classification accuracy.