Multi-level Federated Learning for Industry 4.0 - A Crowdsourcing Approach

Federated learning is one of the emerging areas of research in computer science. It has shown great potential in some application areas and we are witnessing evidence of new approaches where millions or even billions of IoT devices can contribute collectively to achieve a common goal of machine learning through federation. However, existing approaches are primarily suitable for single-task learning with a single objective in a single task owner where it is assumed that the majority of devices contributing to federated learning have a similar design or device type and restrictions. We argue that the true potential of federated learning can only be realised if we have a dynamic and open ecosystem where devices, industrial units, machine manufacturers, non-governmental agencies, and governmental entities can contribute toward learning for multiple tasks and objectives in a crowdsourced manner. In this article, we propose a multi-level framework that shows how federated learning, IoT, and crowdsourcing can come hand-in-hand with each other to make a robust ecosystem of multi-level federated learning for Industry 4.0. This helps build future intelligent applications for Industry 4.0 such as predictive maintenance and fault detection for systems in smart manufacturing units. In addition, we also highlight several use-cases of multi-level federated learning where this approach can be implemented in Industry 4.0. Moreover, if the approach is implemented successfully, besides enhancement in performance it will also help towards a greater common goal e.g. UN Sustainable Goal No 13 i.e. reduction in carbon footprint.