Federated learning provides a promising paradigm to enable network edge intelligence in the future sixth generation (6G) systems. However, due to the high dynamics of wireless circumstances and user behavior, the collected training data is non-independent and identically distributed (non-IID), which causes severe performance degradation of federated learning. To solve this problem, federated learning with non-IID data in wireless networks is studied in this paper. Firstly, based on the derived upper bound of expected weight divergence, a federated averaging scheme is proposed to reduce the distribution divergence of non-IID data. Secondly, to further harmonize the distribution divergence, data sharing is associated with federated learning in wireless networks, and a joint optimization algorithm is designed to keep a sophisticated balance between the model accuracy and the cost. Finally, the simulation results based on a common-used image data set are provided to evaluate the performance of our proposed schemes, which can achieve significant performance gains with a small price of latency and energy consumption.

Based on a hierarchical cloud-fog computingenabled paradigm, fog radio access networks (F-RANs) can provide abundant resource to support the future mobile artificial intelligent services. However, due to the differences of computation and communication capabilities at the cloud computing center, the fog computing based access points (F-APs), and the user devices, it is challenging to propose efficient computation offloading strategies to fully explore the potential of F-RANs. In this paper, we study the design of computation offloading in F-RANs to minimize the total cost with respect to the energy consumption and the offloading latency. In particular, a joint optimization problem is formulated to optimize the offloading decision, the computation and the radio resources allocation. To solve this non-linear and non-convex problem, an iterative algorithm is designed, which can be proved to converge a stationary optimal solution with polynomial computational complexity. Finally, the simulation results are provided to show the performance gains of our proposed joint optimization algorithm.

A key problem of content caching networks is that extra radio resource blocks are consumed to push content objects, which leads to a decline of spectrum efficiency. To solve this problem, a non-orthogonal multiple access-based multicast (NOMA-MC) scheme is proposed in this paper, where pushing and multicasting content objects can be accomplished simultaneously, and thus the spectrum efficiency can be improved significantly. To evaluate the performance of the NOMA-MC scheme, an explicit expression of outage probability is derived, which shows that full diversity gains can be achieved in the single-cell scenario. Moreover, the theoretical results can be extended to the multi-cell scenario by establishing a stochastic geometry-based network model, which show that the NOMA-MC scheme can achieve better performance than the conventional orthogonal multiple access-based multicast scheme. Then, the joint design of power allocation and content matching is studied to enlarge the performance gains of the NOMA-MC scheme, and two distributed optimization algorithms are proposed by solving a hospitals/residents matching problem. Finally, simulation results are provided to verify the analytical results, and also demonstrate the performance gains of the NOMA-MC scheme.

The combination of artificial intelligence and mobile edge computing (MEC) is considered as a promising evolution path of the future wireless networks. As a model-level coordination learning paradigm, federated learning can make full use of the distributed computation resource in the MEC systems, which allows the users to keep their private data locally. However, due to the unreliable wireless transmission circumstances and resource constraints in the MEC systems, both the performance and training efficiency of federated learning cannot be guaranteed. To solve this problem, the optimization design of federated learning in the MEC systems is studied in this paper. First, an optimization problem is formulated to manage the tradeoff between model accuracy and training cost. Second, a joint optimization algorithm is designed to optimize the model compression, sample selection, and user selection strategies, which can approach a stationary optimal solution in a computationally efficient way. Finally, the performance of our proposed optimization scheme is evaluated by numerical simulation and experiment results, which show that both the accuracy loss and the cost of federated learning in the MEC systems can be reduced significantly by employing our proposed algorithm.

Non-orthogonal multiple access (NOMA) technique is capable of improving the efficiency of delivering and pushing contents in wireless caching networks. However, due to the differences of the data volume and the channel condition, the static power control schemes cannot fully explore the potential of NOMA. To solve this problem, dynamic power control for NOMA transmissions in wireless caching networks is studied in this paper, which can be adjusted based on the status of content transmissions. In particular, we focus on minimizing the transmission delay with the considerations of each users transmission deadline and the total power constraint. An iterative algorithm is first proposed to approach the optimal solution of dynamic power control. Then a deep neural network (DNN)-based method is designed to keep a balance between the performance and the computational complexity. Finally, Monte-Carlo simulations are provided for verifications.

Federated learning is a promising technique to implement network intelligence for the sixth generation (6G) communication systems. However, the collected data in wireless networks is non-independent and identically distributed (non-IID), which leads to severe deterioration of model performance. Although various enhanced schemes are proposed, it is still challenging to balance the communication cost and the model performance, due to the scarcity of radio resource for model update in wireless networks. In this paper, an ensemble federated learning paradigm is proposed for handling non-IID data, which is also optimized for its deployment in wireless networks in a cost efficient way. First, the framework of ensemble federated learning is designed. By formulating individual user clusters, intra-cluster federated learning models can be generated to reduce the impact of non-IID data, which can be integrated to adapt to various learning data via model ensemble. Second, the optimization of user cluster formation is studied to improve the performance of ensemble federated learning, which is modeled as a coalition formation game to design a Nash-stable algorithm. Finally, the simulation results on the public data sets are provided to verify the performance gains of our proposed schemes for deploying federated learning with non-IID data in wireless networks.