With the prevalence of emerging artificial intelligence services in next-generation wireless edge networks, Split Federated Learning (SFL), which divides a learning model into server-side and client-side models, has emerged as an appealing technology to deal with the heavy computational burden for network edge clients. However, existing SFL frameworks would frequently upload smashed data and download gradients between the server and each client, leading to severe communication overheads. To address this issue, this work proposes a novel communication-and-computation efficient SFL framework, which allows dynamic model splitting (server- and client-side model cutting point selection) and broadcasting of aggregated smashed data gradients. We theoretically analyze the impact of the cutting point selection on the convergence rate, revealing that model splitting with a smaller client-side model size leads to a better convergence performance and vise versa. Based on the above insights, we formulate an optimization problem to minimize the model convergence rate and latency under the consideration of data privacy via a joint Cutting point selection, Communication and Computation resource allocation (CCC) strategy. To deal with the proposed mixed integer nonlinear programming optimization problem, we develop an algorithm by integrating the Double Deep Q-learning Network (DDQN) with convex optimization methods. Extensive experiments validate our theoretical analyses across various datasets, and the numerical results demonstrate the effectiveness and superiority of the proposed communication-efficient SFL compared with existing schemes, including parallel split learning and traditional SFL mechanisms.

Federated edge learning (FEL) is regarded as a promising distributed machine learning paradigm to reduce transmission latency and resources as well as preserve raw data privacy by collaboratively training local deep learning models across multiple edge devices. However, with the development of artificial intelligence (AI) technologies, the size of neural network models grows exponentially with their parameters to meet variable application requirements, which poses significant challenges to the computation, communication, and memory abilities of edge devices. Existing designs typically focus on either communication or computation efficiency without caring each device’s memory ability. To deal with the above issues, we first introduce a novel model sparse update enabled tiny FEL (TinyFEL) architecture, which terminates the backpropagation early in local model training processes. Therefore, the proposed TinyFEL can reduce local memory occupation and lessen the communication-and-computation burden. Furthermore, we propose a parameter splitting mechanism instead of transmitting the full model, only a part of updated layers of parameters is transmitted for aggregation, which significantly reduced the communication overheads. Thereafter, we develop a communication and computation latency minimization problem to accelerate the training of TinyFEL. To this end, we theoretically analyze the convergence performance of TinyFEL, which unveils the mathematical relationship among sparse update ratio assignment, device selection, and learning performance. Then, a joint sparse update ratio assignment, device selection, and resource allocation strategy is introduced based on the alternating direction method of multipliers (ADMMs) and block coordinate descent (BCD) algorithms. Numerical results indicate that our proposed TinyFEL can reduce training memory occupation by over 40% than the traditional FEL at the cost of negligible accuracy loss.

Edge artificial intelligence (AI) is an emerging solution for pervasive intelligence service in future 6G networks, by learning machine learning (ML) models at network edge. Edge AI typically consists of three processes: sensing, communication, and computation (SC²). Edge devices first collect data samples through the sensing process, then train local models individually through the computation process, and finally update the local models periodically through the communication process to obtain the global model. Federated edge learning (FEEL) is particularly attractive for Edge AI due to its collaborative ML framework and privacy-enhancing feature. However, the research FEEL with SC2 integration remains an open question. On the one hand, there is still a lack of theoretical insight into the learning performance that is jointly influenced by the processes of SC2. On the other hand, the performance evaluation is another challenge for the proposed SC2-FEEL, which further poses the difficulties in design of efficient resource allocation. To address these issues, an SC2 integrated FEEL (SC2-FEEL) is investigated in this article, where the processes of SC2 are jointly considered and the over-the-air computation (AirComp) technique is employed for a communication-efficient model aggregation. First, theoretical analyses are conducted, which reveals both the sample sensing strategy and the AirComp-induced communication error significant affect the learning performance of SC2-FEEL. Then, we further formulate a latency and energy consumption minimization problem with learning performance guaranteed based on the theoretical results, which is mixed integer nonlinear programming (MINLP) and dynamic programming. To deal with this problem, we propose a joint SC2 resource optimization strategy with low complexity based on the block coordinate update and Lyapunov optimization framework. Extensive simulation results are provided to validate our theoretical analysis, and demonstrate the effectiveness of developed algorithm.

Communication and energy efficiencies are two crucial objectives in the pursuit of edge intelligence in 6G networks, and become increasingly important given the prevalence of large model training. Existing designs typically focus on either communication efficiency or energy efficiency due to the fact that improving one objective generally comes at the expense of the other. Over-the-air federated learning (OTA-FL) has recently emerged as a promising approach to enhance both efficiencies through an integrated communication and computation design. Nevertheless, most previous studies on OTA-FL only consider scenarios where the dataset for the entire FL procedure is collected and available prior to training. In real-world applications, devices continuously collect new data in an online manner. This underscores the significance of sample collection through sensing in a practical FL pipeline. We propose to integrate sensing with communication and computation into a joint design to further boost the communication-and-energy efficiencies of OTA-FL. Specifically, we consider a training latency and energy consumption minimization problem with performance guarantees. To this end, we first derive an average training error (ATE) metric to quantify convergence performance. Then, a joint sensing, communication and computation resource allocation strategy is developed based on a deep reinforcement learning (DRL) algorithm that nests convex optimization with a deep Q-network. Extensive experiments are conducted to validate our theoretical analysis, and demonstrate the effectiveness of the proposed design for communication-and-energy efficient FL.

Integrated sensing and communication (ISAC) within the unlicensed millimeter-wave (mmWave) frequency bands has been emerged as a pivotal technology in the next generation wireless communication era. However, the interference management issue between sensing and communication becomes much severe due to the absence of centralized scheduling function of the widely existed WiGig networks with the IEEE 802.11ay protocol in the unlicensed mmWave bands. In this way, we aim to investigate an efficient ISAC scheme for the promising WiGig networks via embedding radar pulses into the IEEE 802.11ay beamforming training (BFT) period. Particularly, we transmit both radar pulses and communication signals by exploiting the sector-level sweep of the WiGig network. Since there is a trade-off between radar detection/sensing and mmWave communication under diverse resource assignments, we propose a joint BFT and energy allocation strategy to find an achievable balance. Numerical results validate the effectiveness of the proposed scheme.

Federated multi-task learning (FMTL) is a promising technology to deal with the severe data heterogeneity issue in federated learning (FL), where each client learns individual models locally and the server extracts similar model parameters from the tasks to keep personalization for models of clients. Hence, it is essential to precisely extract the model parameters shared among tasks. On the other aspect, the limitation of communication resources would also restrict the model transmission, and thus influence the FMTL performance. To address the above issues, we propose a novel FMTL with Sparse Sharing (FedSS) mechanism that allows clients to share model parameters dynamically according to diversified model structures under limited communication resources. Particularly, we present an adaptive quantization approach for task relevance, which serves as a metric to evaluate the extent of model sharing across tasks. The objective function is formulated to minimize the model transmission latency while ensure the FMTL learning performance via a joint bandwidth allocation and client selection strategy. Closed-form expressions for the optimal client selection and bandwidth allocation are derived based on a alternating direction method of multipliers (ADMM) algorithm. Numerical results show that the proposed FedSS outperforms the benchmarks, and achieves efficient communication performance.

In this paper, we consider a ground terminal (GT) to an unmanned aerial vehicle (UAV) wireless communication system where data from GTs are collected by an unmanned aerial vehicle. We propose to use the ground terminal-UAV (G-U) region for the energy consumption model. In particular, to fulfill the data collection task with a minimum energy both of the GTs and UAV, an algorithm that combines optimal trajectory design and resource allocation scheme is proposed which is supposed to solve the optimization problem approximately. We initialize the UAV’s trajectory firstly. Then, the optimal UAV trajectory and GT’s resource allocation are obtained by using the successive convex optimization and Lagrange duality. Moreover, we come up with an efficient algorithm aimed to find an approximate solution by jointly optimizing trajectory and resource allocation. Numerical results show that the proposed solution is efficient. Compared with the benchmark scheme which did not adopt optimizing trajectory, the solution we propose engenders significant performance in energy efficiency.