Federated learning (FL) is promising in enabling large-scale model training by massive devices without exposing their local datasets. However, due to limited wireless resources, traditional cloud-based FL system suffers from the bottleneck of communication overhead in core network. Fueled by this issue, we consider a hierarchical FL system and formulate a joint problem of edge aggregation interval control and resource allocation to minimize the weighted sum of training loss and training latency. To quantify the learning performance, an upper bound of the average global gradient deviation, in terms of the edge aggregation interval, the training latency, and the number of successfully participating devices, is derived. Then an alternative problem is formulated, which can be decoupled into an edge aggregation interval control problem and a resource allocation problem, and solved by an iterative optimization algorithm. Specifically, given the resource allocation strategy, a relaxation and rounding method is proposed to optimize the edge aggregation interval. The problem of resource allocation including training time allocation and bandwidth allocation is solved separately based on the convex optimization theory. Simulation results show that the proposed algorithm, compared to the benchmarks, can achieve higher learning performance with lower training latency, and is capable of adaptively adjusting the edge aggregation interval and the resource allocation strategy according to the training process.

Federated learning (FL) enables clients to collaboratively learn a shared task while keeping data privacy, which can be adopted at the edge of wireless networks to improve edge intelligence. In this letter, we aim to minimize the training latency of a wireless FL system for a given training loss by client scheduling. Instead of assuming that the prior information about wireless channel state and local computing power of the clients is available, we consider a more practical scenario without knowing the prior information. We first reformulate the client scheduling problem as a multi-armed bandit program and then propose an online scheduling scheme based on ϵ-greedy algorithm to achieve a tradeoff between exploration and exploitation. In addition, the proposed client scheduling scheme reduces the number of training rounds and the time interval per round simultaneously by jointly considering the significance of local updates and delay issue of each client. Simulation results show that in the case of non-independent and identically distributed data, the proposed scheme can save half the training time compared to the scheme which only considers the significance of local updates, and can improve more than 20% test accuracy compared to the scheme which only considers the time consumption per round of each client.

Federated learning (FL) framework enables user devices to collaboratively train a global model based on their local data sets without privacy leak. However, the training performance of FL is degraded when the data distributions of different devices are incongruent. Fueled by this issue, we consider a clustered FL (CFL) method where the devices are divided into several clusters according to their data distributions and are trained simultaneously. Convergence analysis is conducted, which shows that the clustered model performance depends on cosine similarity, device number per cluster, and device participation probability. Besides, to quantify the training performance, the utility of clustered model training is defined based on the analysis results. Then, aiming at optimizing the system utility, a joint problem of resource allocation and device clustering is formulated, which is solved by decoupling it into two subproblems. First, given the results of device clustering, a low-complexity iterative algorithm based on the convex optimization theory is proposed to make the bandwidth allocation and the transmit power control. Then, according to the individual stability, a coalition formation algorithm is proposed for the device clustering. Finally, the real-data experiments on the classification tasks (e.g., MNIST, CIFAR-10, and CIFAR-100) validate the results of convergence analysis and advantages of the proposed algorithm in terms of the test accuracy.

Power Internet of Things (PIoT) is a promising solution to meet the increasing electricity demand of modern cities, but real-time processing and analysis of huge data collected by the devices is challengeable due to limited computing capability of devices and long distance from the cloud center. In this paper, we consider the edge computing assisted PIoT where the computing tasks of the devices can be either processed locally by the devices, or offloaded to edge servers. Aiming to maximize the long-term system utility which is defined as a weighted sum of reduction in latency and energy consumption, we propose a novel task offloading algorithm based on deep reinforcement learning, which jointly optimizes task scheduling, transmit power of the PIoT devices, and computing resource allocation of the edge servers. Specifically, the task execution on each edge server is modeled as a queuing system, in which the current queue state may affect the waiting time for the next tasks. The transmit power and computing resource allocation is first optimized, respectively, and then a deep Q-learning algorithm is adopted to make task scheduling decisions. Numerical results show that the proposed method can improve the system utility.

The Industrial Internet of Things (IIoT) is emerging as a promising technology that can accelerate the application of industrial intelligence to smart factories. Because of the sensitive nature of user data, federated learning (FL) which performs distributed machine learning while preserving data privacy, is leveraged to meet the accuracy and privacy requirements of IIoT end devices/clients. However, the unreliable communications in IIoT may result in possible single-point failures in the typical single-server FL framework, thereby negatively affecting the training efficiency. In this article, we study on the client scheduling problem in a multiserver FL framework for the communication reliability and training efficiency improvement. Specifically, we focus on a semi-decentralized FL (SD-FL) framework, where edge servers and clients collaborate to train a shared global model through unreliable intracluster model aggregation and intercluster model consensus because of the model transmission error in client–server and server–server communication. Then, a client–server association optimization problem is formulated, with the objective of minimizing the global training loss. Resorting to the convergence analysis of SD-FL, the original problem is simplified and transformed into an integer nonlinear programming problem to guide us to design a high-efficiency client scheduling scheme. Finally, experimental results show that the proposed scheme significantly outperforms the baselines in terms of the test accuracy and training loss.