Federated learning (FL) is a distributed deep learning method that enables multiple participants, such as mobile and IoT devices, to contribute a neural network while their private training data remains in local devices. This distributed approach is promising in the mobile systems where have a large corpus of decentralized data and require high privacy. However, unlike the common datasets, the data distribution of the mobile systems is imbalanced which will increase the bias of model. In this paper, we demonstrate that the imbalanced distributed training data will cause an accuracy degradation of FL applications. To counter this problem, we build a self-balancing FL framework named Astraea, which alleviates the imbalances by 1) Z-score-based data augmentation, and 2) Mediator based multi-client rescheduling. The proposed framework relieves global imbalance by adaptive data augmentation and downsampling, and for averaging the local imbalance, it creates the mediator to reschedule the training of clients based on Kullback-Leibler divergence (KLD) of their data distribution. Compared with FedAvg, the most popular FL framework, Astraea shows +4.39% and +6.51% improvement of top-1 accuracy on the imbalanced EMNIST and imbalanced CINIC-10 datasets, respectively. Meanwhile, the communication traffic of Astraea is reduced by 75% compared to FedAvg.

Federated Learning (FL) enables the multiple participating devices to collaboratively contribute to a global neural network model while keeping the training data locally. Unlike the centralized training setting, the non-IID, imbalanced (statistical heterogeneity) and distribution shifted training data of FL is distributed in the federated network, which will increase the divergences between the local models and the global model, further degrading performance. In this paper, we propose a flexible clustered federated learning (CFL) framework named FlexCFL, in which we 1) group the training of clients based on the similarities between the clients’ optimization directions for lower training divergence; 2) implement an efficient newcomer device cold start mechanism for framework scalability and practicality; 3) flexibly migrate clients to meet the challenge of client-level data distribution shift. FlexCFL can achieve improvements by dividing joint optimization into groups of sub-optimization and can strike a balance between accuracy and communication efficiency in the distribution shift environment. The convergence and complexity are analyzed to demonstrate the efficiency of FlexCFL. We also evaluate FlexCFL on several open datasets and made comparisons with related CFL frameworks. The results show that FlexCFL can significantly improve absolute test accuracy by $+10.6\%$ on FEMNIST compared with FedAvg , $+3.5\%$ on FashionMNIST compared with FedProx , $+8.4\%$ on MNIST compared with FeSEM , $+4.7\%$ on Sentiment140 compare with IFCA . The experiment results show that FlexCFL is also communication efficient in the distribution shift environment.

The Internet of Things (IoT) has emerged as one of the key features of the next-generation wireless networks, where timely delivery of status update packets is essential for many real-time IoT applications. Age of Information (AoI) is a new metric to measure the freshness of update. The reduction of the violation probability that AoI of status updates exceeds a given age constraint is of great significance for guaranteeing the information freshness in IoT systems. By modeling the IoT networks as M/M/1 and M/D/1 queuing systems, this work focuses on characterizing the violation probability of peak AoI and AoI in IoT systems, where a sensor delivers updates to a monitor under M/M/1 and M/D/1 queues with first-come-first-served policy. From a time-domain perspective, we explore the correlation between interdeparture time and system time, by which the closed-form expressions of peak AoI distribution and the violation probability for any AoI constraint are derived. The obtained results induce accurate characterizations for probability distribution functions of peak AoI and AoI. Consequently, accurate characterizations of average AoI and the variance of AoI are obtained. Then, for peak AoI and AoI, the optimal generation rate of the status update that induces the minimal violation probability is also found. The numerical results show that the optimal update rate can significantly reduce the AoI violation probability for a wide range of AoI constraints. The theoretical findings and predictions are verified by numerical simulation results as well as provide guidance for the design of IoT networks.

The large amount of data amongst billions of devices deployed for Industrial Internet of Things (IIoT) cause a massive energy demands. Driven by the pursuit of green communication, this paper presents space-reserved cooperative caching in the 5th generation (5G) mobile heterogeneous networks for IIoT, where the caching space in base stations (BSs) is largely utilized to store the fixed caching data (FCD) prefetched from the servers, and the reserved space is to temporarily store the data to be wirelessly transmitted (DWT). With the constraint that the quality of service is guaranteed, we propose an algorithm to obtain the optimal proportion between the two parts of the caching space for the purpose of reducing the average energy consumption. Simulation results verified that the proposed caching scheme is more efficient than the conventional one with respect to the average energy consumption.

Ultra-dense networks (UDNs), which can provide extremely high throughput and data rates, have been considered as one of the key techniques for the fifth generation mobile networks. However, it may cause severe inter-cell interference and significant energy consumption due to numerous base stations (BSs) being randomly deployed. To mitigate the interference and boost energy efficiency (EE) of the UDN effectively, we propose a cluster-based energy-efficient resource allocation scheme in this paper. The proposed scheme has two stages: clustering stage and resource allocation stage. In clustering stage, we use a modified K-means algorithm in BS-clustering process to dynamically adjust the number of BS-clusters based on the density of BSs. Then, in each BS cluster, we divide user equipments (UEs) into multiple UE-groups with minimum intra-cluster interference. In this way, the complexity of resource allocation can be greatly reduced. While in resource allocation stage, we design a two-step resource blocks assignment algorithm and an iterative energy efficient power allocation algorithm based on a non-cooperative game. Furthermore, we implement simulations under the realistic broadband channel propagation conditions and the simulation results show that the proposed approach can effectively mitigate the interference and improve the EE of UDN.

In future mobile networks, more and more machine-type communication (MTC) devices with different service requirements will be deployed. To meet the massive access needs of MTC, this paper develops a cluster-based congestion-mitigating access scheme (CCAS), with aim to mitigate the severe collision of MTC devices (MTCDs) that access to the base station concurrently. To this end, we first design a modified spectral clustering algorithm to group MTCDs into different clusters based on their locations and service requirements. Then, a device called MTC gateway (MTCG) is chosen by two steps to assist transmitting data for MTCDs in each cluster. In the data transmission process, MTCG is in charge of aggregating packets generated by MTCDs in a cluster and forwarding them to BS when the number of buffered packets reaches a certain threshold. To model the aggregation and forwarding process of each MTCG, we use queuing theory to analyse the access performance in terms of collision probability and access delay. In addition, we also implement simulations to further validate the accuracy of our analytical model and the effectiveness of CCAS. Numerical results, which are consistent with the theoretical values, show that the proposed CCAS can significantly decrease collision probability, and increase the number of successfully received packets of the system without increasing average access delay.