Federated learning has been gaining adoption in mobile and cloud computing as a mechanism for decentralizing both computation and trust when building machine learning models from user data. Proposals for using federated learning on IoT devices have been emerging, but even though federated learning could provide valuable insights from IoT devices faster and more efficiently, its adoption has been lackluster.
We argue that this slow adoption is due to two challenges. First, federated learning methods are designed for cloud servers and mobile processors, which are significantly more powerful than typical IoT devices. Second, the level of trust we can place in code executed on IoT devices is significantly lower than on mobile devices or cloud servers.
We aim to overcome these challenges by redesigning the relevant algorithms to run more efficiently on IoT hardware. As an additional benefit of this redesign, the resulting applications have a small memory footprint and low communication overhead, which makes it easier to secure their execution and make them trusted on the IoT device.
More information: Project webpage