A Federated Actor‐Critic Framework for UAV ‐Assisted Mobile Edge Computing With Prospect‐Theoretic Optimization

Integrating mobile edge computing (MEC) with resource‐constrained unmanned aerial vehicles (UAVs) enables flexible, on‐demand resource provisioning. Data security and privacy, however, remain critical challenges in UAV swarm networks, largely because of the uncertainties inherent in resource sharing, and it is essential that all ground devices (GDs) receive equitable and secure opportunities to offload tasks to UAVs. To address these challenges, we propose a federated learning (FL)‐based mechanism that incorporates differential privacy to balance transmission efficiency and security in UAV swarm networks. Specifically, we jointly optimize the differential privacy parameters, transmit power, and UAV trajectories within a prospect theory‐based Asynchronous Advantage Actor‐Critic (A3C) framework, subject to path bandwidth consumption, UAV flight constraints, and limited communication resources. In this framework, UAVs collaboratively train model parameters under a differentially private federated learning protocol, and stochastic geometry tools are used to derive the success probability of model parameter transmissions to the central aggregator in each global aggregation round. Extensive simulations show that the proposed mechanism preserves data security and privacy in UAV swarm networks while achieving minimal system delay.