Integrated Simulation of Federated Learning for Large Scale Intelligent Transportation Systems

Federated Learning (FL) is a decentralized learning paradigm that is expected to play a substantial role in both the network management of 5G and beyond protocols and smart city applications like Intelligent Transportation System (ITS). The massive scale and decentralized nature of FL makes them depend on Vehicular Ad-hoc Networks (VANETs) to support model exchanges between the FL participants. However, traditional FL simulators are not designed to model the communication behavior of a VANET. We address this gap by proposing an integrated simulation approach to facilitate combined studies of FL and VANET. We describe the necessary design changes to common simulation components and implement a sample simulator by extending a simplified VANET simulator. We demonstrate the need of such integrated simulation, illustrating that typical VANET effects substantially impact FL training – and that FL training substantially impacts ITS infrastructure. We also demonstrate the fidelity and performance of our approach by comparing it with a traditional FL simulator. Finally, we discuss the applications and limitations of our proposed approach.