| Popis: |
Content caching has been considered as an effective way to offload contents at network edge in order to alleviate backhaul load. Recently, federated learning (FL) based edge caching has gained a lot of popularity due to its prominent features of data privacy, distributed mode of operation, and scalability. However, these FL-based schemes ignore the behavior of the communication channels during the federated weight averaging procedure. In this paper, we introduce a novel robust federated learning-based content caching approach for fog radio access networks (F-RANs) that mitigates the effect of communication channel noise. In our proposed robust FL approach, each cell employs a deep neural network (DNN)-based model to predict users’ future files rating score based on user and file contextual information and shares its learned weights to the fog server. The fog server is responsible for global weight averaging. Prior to FL weight averaging fog sever feed incoming local model weights to a generative adversarial neural network (GANs) model which differentiates between noisy and actual federated weights, and passes only actual weights based on the distribution of the weight matrices. Extensive simulations have been carried out to validate the performance of our proposed approach. Results show that the GAN-aided federated model yields 23.1% more prediction accuracy as compared to the federated noisy model without GANs based noise mitigation. |
