Resource Allocation and Optimization for Federated Learning in Vehicular Edge Caching

Edge caching enables vehicles to access content directly from roadside units (RSUs), which can effectively improve the user experience in the current era of content explosion. However, due to the limited cache capacity of RSUs, it is of great importance to cache hotspot content on RSUs and improve the hit rate of cached content. Many researchers employ federated learning to train models for content prediction, which protects user privacy and utilizes user information data. However, few studies have considered the distribution and aggregation of federated learning models in the context of restricted RSU communication resources. In this paper, we investigate the use of federated learning for edge caching and propose a Resource Allocation for Federation Learning in Content Caching (RAFLCC) scheme. The problem is modeled as an energy minimization problem under time constraints. Based on the modeled problem, a reasonable allocation of resources is achieved, allowing more vehicles to participate in federated learning training. The experimental results demonstrate that RAFLCC outperforms other baseline caching schemes.