This experiment aims at providing a federated deep learning solution which allows multiple DIHs to collaborate on an AI tool by locally training a neural network on their data and then combining these into one larger, more efficient solution.
To do so, the solution is initially developed as a proof of concept using an example dataset and deep learning model that can later be updated to become suitable for more applicable use cases, such as, but not limited to, those dealing with manufacturing.
The experiment also investigates the possibility of setting up this kind of federated deep learning solution and displays it as a reusable component by DIHs. The solution itself is built based on IDS and makes use of a dataspace for DIHs which will be deployed in a later stage of the experimentation. Moreover, the DIHs connected to the dataspace can deploy their own connectors to local partners so to to involve them in the platform.
Providing an AI Testing and Experimental Facility in Manufacturing through DIHs enables combining different dispersed technologies and datasets in Europe. Therefore, enabling use-cases such as AI training on local data and then combining the AIs in a single neural network results in a more powerful and practically usable solution that couldn’t be obtained otherwise. This merges together the ideas of data sovereignty and federalization - which already exist in Europe - with the technical reality which demonstrates how larger datasets are needed to keep up with the global AI development speed. To this extent, a federated deep learning solution would allow European DIHs to acquire a key position in the field of monitoring AI developments and would enable DIHs' partners to gain a key competitive advantage in the development and usage of AI applications.