The currently ongoing revolution in machine learning (ML) and AI is largely driven by the wide availability of data. The rapidly developing IoT is intimately connected to this process: advanced ML methods improve IoT, which drives further progress by providing large amounts of new data. This presents several challenges for research: IoT data are very heterogeneous, and the devices that form the IoT have different computational and communication capabilities. In addition, since IoT is intimately embedded into people's lives, the data have the potential to seriously undermine privacy of the individuals if mishandled.
We aim to solve key challenges facing IoT by providing an easy-to-use modelling framework that is scalable, enables the use of powerful probabilistic models to account for complex dependencies in the data, and has a strong, built-in privacy protection. We will demonstrate our solution with prototype applications on an IoT platform.