Privacy-Preserving Edge Machine Learning on Internet of Things (IoT) Systems with Extreme Resource Constraints

Today, we are witnessing a trend towards the Internet of Intelligent Things (IoT). The opportunities brought about by such systems can create a game-changing shift in our society.

One of the main application domains, which can revolutionize our healthcare system, is real-time continuous health monitoring using wearable and mobile-health technologies, for early detection/prediction of health pathologies on a personalized basis.

Such opportunities have been enabled mainly as a result of the synergies between IoT technologies and artificial intelligence, through the so-called big data acquired by the IoT devices and analyzed by artificial intelligence and machine learning techniques.

One of the main challenges in fully exploiting these recent opportunities is certainly the security and privacy concerns of the users, particularly in the medical and health domain.

Such privacy concerns are due to the inherent sensitivity of personal health/medical data, which needs to be analyzed for decision-making by machine-learning algorithms.

The majority of the privacy-preserving machine-learning literature, however, focuses on high-end mobile devices such as smartphones, which by no means represent the typical IoT and mobile-health sensors/devices with extreme resource constraints.

In this proposal, our goal is to enable privacy-preserving learning and inference under extreme resource constraints for IoT and mobile-health technologies.