On Addressing Heterogeneity in Federated Learning for Autonomous Vehicles Connected to a Drone Orchestrator

In this paper we envision a federated learning (FL) scenario in service of amending the performance autonomous road vehicles, through drone traffic monitor (DTM), that also acts as an orchestrator. Expecting non-IID data distribution, focus on issue accelerating particular class critical object (CO), may harm nominal operation vehicle. This can be done proper allocation wireless resources for addressing learner and heterogeneity. Thus, propose reactive method resources, happens dynamically each FL round, is based learner’s contribution to general model. addition this, explore use static methods remain constant across all rounds. Since expect partial work from learner, FedProx algorithm, task computer vision. For testing, construct distribution MNIST FMNIST datasets among four types learners, scenarios represent quickly changing environment. The results show proactive measures are effective versatile at improving system accuracy, CO when underrepresented network. Furthermore, experiments tradeoff between intensity resource efforts. Nonetheless, well adjusted local optimizer allows even better overall particularly using deeper neural network (NN) implementations.