Edge Caching with Federated Unlearning in Cluster-Centric Small Cell Networks
Lingrui Cui, Pengfei Wang, Yuqi Han · 2024
Edge content caching based on federated learning has emerged as a promising technology to address the increasing personalized demands of compute-intensive and latency-sensitive applications within small cell network scenarios. Existing content caching schemes often struggle to accommodate personalized caching requirements while processing invalid data such as poisoned and stale data. To address these challenges, this paper proposes a federated unlearning-enabled small cell network caching paradigm (FedSCC). Firstly, we introduce a personalized federated learning caching scheme based on user classification. This scheme categorizes users according to their preferences and caches content tailored to the preferences of corresponding user classes in small base stations, enhancing cache hit rates and reducing service latency. Secondly, we present a content caching algorithm based on federated unlearning to further improve the caching paradigm's performance. By mitigating the adverse effects of invalid data on the global model and notably reducing service latency, this algorithm significantly enhances cache performance. Experimental results demonstrate that in small-cell network scenarios characterized by personalized caching demands, FedSCC can elevate cache hit rates by at least 21.73% compared to baselines. Additionally, cache hit rates exhibit a significant improvement of 27.05% in experiments involving invalid data.