Coordinated Scheduling and Decentralized Federated Learning Using Conflict Clustering Graphs in Fog-Assisted IoD Networks

Despite the advantages of fog-assisted internet of drones (IoD) networks for federated learning (FL) model aggregations, it is restricted by the limited battery capacity of drones and wireless channel conditions between a fog and drones. In this paper, we address the deployment problem of decentralized FL and the secrecy rate maximization problem in fog-assisted IoD networks. In particular, we leverage the overlapped zones of drones to spread their local aggregated models to the network. As such, the global model aggregation is achieved without any participation from the fog. Then, aiming to secure the proposed decentralized FL-based framework, we consider the secrecy rate maximization problem, which can be addressed by scheduling users to drones and their available radio resource blocks (RRBs). This coordinated scheduling problem of users-drones/RRBs is NP-hard. To efficiently solve it using graph theory techniques, we construct a centralized conflict clustering graph and reformulate the considered scheduling problem as a maximum-weight independent set (MWIS) problem in the constructed conflict graph. To alleviate the need for the coordinated scheduling policy at the fog, we propose to solve the coordinated scheduling problem in a distributed manner. We conduct extensive simulations to verify the effectiveness of the proposed schemes over existing schemes. Selected numerical results show that the proposed distributed scheduling schemes provide similar performances in terms of secrecy rate and energy consumption as compared with the centralized scheme, with reasonable amount of information exchanges among the drones.