Developing and Deploying Federated Learning Models in Data Spaces: Smart Truck Parking Reference Use Case

Earlier work proposed a reference use case and data space architecture for smart truck parking and positioned future use of federated learning for competition-, privacy sensitive data sharing. However, there is limited research regarding the deployment of federated learning in data spaces. Extending earlier work, this paper documents the results of experimental development of a federated learning model for smart truck parking and its instantiation in a data space infrastructure. Two iterations were carried out to assess the development of a federated learning model and deployment in a data space environment for the smart truck parking use case. First, a data space infrastructure was instantiated, containing a federated learning orchestrator, connectors with data apps, and a metadata broker. Second, a prototype was developed on top of the metadata broker to support the provisioning of the required data space components to the involved participants. Taken together, the experimental development related to the smart truck parking case provides initial support for the suitability of federated learning in a data space environment and contributes to better understanding of the potential use, technical feasibility, required efforts, and practical implications. From a practical perspective, the study provides interested scholars and software developers access to a reference implementation. The current study is limited to one federated learning model and deployment in a small data space environment. Future work may contribute to comparing multiple federated learning models and evaluation in an operational data space.