With the advancement of sixth generation (6G) communication technology and artificial intelligent (AI), Electric Intelligent Vehicles (EIVs) have led to a rapid increase in time-bounded and ultra-reliable tasks. However, the growing number of EIVs imposes a high computational complexity on traditional learning-based offloading schemes. This paper proposes a mean field deep reinforcement learning (MFDRL) framework to optimize task offloading decisions aiming to minimize energy consumption for each EIV. Specifically, we first formulate the energy minimization problem as a Markov game while considering the mobility of EIVs, the limited computation resources of EIVs, and the task delay constraints. To reduce the computational complexity of traditional DRL algorithms in deriving the Nash equilibrium in scenarios with a large number of players, we develop a novel DRL algorithm that incorporates mean field theory (MFT), which approximates the joint action space by mean field action. Unlike existing learning-based task offloading approaches, the proposed framework enables each EIV to respond only to the average behavior of the other EIVs, rather than requiring complete knowledge of their individual offloading policies. In this way, the challenge of dimensionality explosion can be effectively mitigated. Simulation results reveal that: 1) the proposed algorithm converges faster than benchmark algorithms, and can achieve superior performance to two benchmark schemes in terms of delay and energy consumption and 2) the proposed algorithm can achieve better scalability than the multi-agent deep deterministic policy gradient (MADDPG) algorithm as the number of TEIVs increases.

Deploying trustworthy AI systems requires principled uncertainty quantification. Conformal prediction (CP) is a widely used framework for constructing prediction sets with distribution-free coverage guarantees. In many practical settings, including healthcare, finance, and mobile sensing, the calibration data required for CP are distributed across multiple clients, each with its own local data distribution. In this federated setting, data can often be partitioned into, potentially overlapping, groups, which may reflect client-specific strata or cross-cutting attributes such as demographic or semantic categories. We propose group-conditional federated conformal prediction (GC-FCP), a federated extension of conditional conformal calibration for a target mixture over prespecified groups. GC-FCP constructs mergeable, atom-stratified coresets from local calibration scores, enabling compact aggregation at the server when the number of active atoms is moderate. Experiments on synthetic and real-world datasets validate the performance of GC-FCP compared to centralized calibration baselines. The code of our work can be found at https://github.com/HaifengWen/GC-FCP.

For modern artificial intelligence (AI) applications such as large language models (LLMs), the training paradigm has recently shifted to pre-training followed by fine-tuning. Furthermore, owing to dwindling open repositories of data and thanks to efforts to democratize access to AI models, pre-training is expected to increasingly migrate from the current centralized deployments to federated learning (FL) implementations. Meta-learning provides a general framework in which pre-training and fine-tuning can be formalized. Meta-learning-based personalized FL (meta-pFL) moves beyond basic personalization by targeting generalization to new agents and tasks. This paper studies the generalization performance of meta-pFL for a wireless setting in which the agents participating in the pre-training phase, i.e., meta-learning, are connected via a shared wireless channel to the server. Adopting over-the-air computing, we study the trade-off between generalization to new agents and tasks, on the one hand, and convergence, on the other hand. The trade-off arises from the fact that channel impairments may enhance generalization, while degrading convergence. Extensive numerical results validate the theory.

To support the unprecedented growth of the Internet of Things (IoT) networks, tremendous data need to be collected by the IoT devices and processed. Due to the limited computing capabilities of IoT devices, part of collected data need to be offloaded to the edge server for further processing. The conventional designs separately considering sensing, communication and computation processes lead to severe wastes of radio, energy, and computation resources. To overcome this drawback, an integrated sensing-communication-computation (ISCC) design is proposed in this paper, which aims at realizing energy efficient data processing by jointly determining the data offloading ratio together with the sensing and offloading rates according to the central processing units (CPU) profiles of IoT devices and servers. To deal with the resultant non-trivial optimization problem, the case with full-utilization of the server's CPU resource is considered, where the data offloading ratio is directly determined by the server's CPU profile, and the string-pulling (SP) algorithms are proposed to obtain the optimal sensing and offloading rates. As for the case with under-utilization of the server's CPU resource, a master-slave solving approach is proposed, where the master problem for optimizing the data offloading ratio can be solved by sub-gradient method, while the slave problem for optimizing the sensing and offloading rates can be solved by SP algorithms. The above solving approaches are further modified to account for the case with sensing busy time at which no data can be sensed by the IoT device. Simulations are conducted to verify the effectiveness of the proposed algorithms.

Federated Learning (FL) is a distributed learning paradigm that preserves privacy without raw data exchange though, it is still vulnerable to dedicated privacy attacks. In wireless FL settings, where analog over-the-air computing (AirComp) is leveraged to enable scalable and communication-efficient training, referred to as AirFL, prior arts harnessed inherent channel noise thereof to strengthen privacy preserving, formally characterized by differential privacy (DP). However, these studies often impose strong assumptions on, e.g., channel models, loss functions or level of DP. In this work, we investigate user-level DP guarantee for AirFL systems with multi-antenna base stations (BSs). Specifically, we derive a convergent DP bound that has no reliance on artificial noise (AN) under a mild assumption of bounded-domain model parameters. To characterize tradeoffs between achievable DP and learning performance, we then analyze the convergence for DP-guaranteed AirFL with general smooth and non-convex loss functions. Next, we jointly optimize receive beamforming and power allocations to further improve on such trade-offs, revealing explicit conditions in which DP is achievable without any compromise to training, followed by validation through numerical experiments.

With the emergence of simultaneous localization and communication (SLAC), it becomes more and more attractive to perform angle of departure (AoD) estimation at the receiving Internet of Thing (IoT) user end for improved positioning accuracy, flexibility and enhanced user privacy. To address challenges like a large number of real-time measurements required for latency-critical applications and enormous data collection for training deep learning models in conventional AoD estimation methods, we propose in this letter an unsupervised learning framework, which unifies training for both deterministic maximum likelihood (DML) and stochastic maximum likelihood (SML) based AoD estimation in multiple-input single-output (MISO) downlink (DL) wireless transmissions. Specifically, under the line-of-sight (LoS) assumption, we incorporate both the received signals and pilot-sequence information, as per its availability at the DL user, into the input of the deep learning model, and adopt a common neural network architecture compatible with input data in both DML and SML cases. Extensive numerical results validate that the proposed unsupervised learning based AoD estimation not only improves estimation accuracy, but also significantly reduces required number of observations, thereby reducing both estimation overhead and latency compared to various benchmarks.

A novel mobile edge generation (MEG) framework is proposed for low-latency image generation on mobile devices. Exploiting a latent diffusion model (LDM) distributed across the edge server (ES) and the user equipment (UE), only low-dimension features need to be transmitted for creating artificial intelligence generative content (AIGC). Two novel modules, namely dynamic diffusion and feature merging, are conceived to compress the diffusion model and transmitted features, respectively. By jointly optimizing compression rates of denoising steps and feature merging, the image quality maximization problem is formulated subject to latency and energy consumption constraints. To address this problem in dynamic channel conditions, a low-complexity compression protocol is developed. First, a backbone LDM architecture is learned by offline distillation to support various compression options. Then, compression rates are predicted in online environment specific to channel and task features. To solve the resultant constrained Markov Decision Process (MDP), a constrained variational policy optimization (CVPO) based MEG algorithm, MEG-CVPO, is further developed to learn constraint-guaranteed optimization. Numerical results demonstrate that: 1) The proposed framework improves image distortions while reducing over 40% latency compared to conventional generation schemes. 2) MEG-CVPO stringently guarantee constraints and realizes a flexible trade-off between generation qualities and overheads. Code is available at https://github.com/xiaoxiaxusummer/LowLatencyMEG

This letter proposes a novel user localization and channel estimation framework for pinching-antenna systems (PASS), where pinching antennas are grouped into subarrays on each waveguide to cooperatively estimate user/scatterer locations, thus reconstructing channels. Both single-waveguide (SW) and multi-waveguide (MW) structures are considered. SW consists of multiple alternatingly activated subarrays, while MW deploys one subarray on each waveguide to enable concurrent subarray measurements. For the 2D scenarios with a fixed user/scatter height, an orthogonal matching pursuit-based geometry-consistent localization (OMP-GCL) algorithm is proposed, which leverages inter-subarray geometric relationships and compressed sensing for precise estimation. Theoretical analysis on Cramér-Rao lower bound (CRLB) demonstrates that: 1) The estimation accuracy can be improved by increasing the geometric diversity through multi-subarray deployment; and 2) SW provides a limited geometric diversity within a 180^∘ half space and leads to angle ambiguity, while MW enables full-space observations and reduces overheads. The OMP-GCL algorithm is further extended to 3D scenarios, where user and scatter heights are also estimated. Numerical results validate the theoretical analysis, and verify that MW achieves centimeter- and decimeter-level localization accuracy in 2D and 3D scenarios with only three waveguides.

Federated Learning (FL) is a distributed learning paradigm that preserves privacy by eliminating the need to exchange raw data during training. In its prototypical edge instantiation with underlying wireless transmissions enabled by analog over-the-air computing (AirComp), referred to as over-the-air FL (AirFL), the inherent channel noise plays a unique role of frenemy in the sense that it degrades training due to noisy global aggregation while providing a natural source of randomness for privacy-preserving mechanisms, formally quantified by differential privacy (DP). It remains, nevertheless, challenging to effectively harness such channel impairments, as prior arts, under assumptions of either simple channel models or restricted types of loss functions, mostly considering (local) DP enhancement with a single-round or non-convergent bound on privacy loss. In this paper, we study AirFL over multiple-access fading channels with a multi-antenna base station (BS) subject to user-level DP requirements. Despite a recent study, which claimed in similar settings that artificial noise (AN) must be injected to ensure DP in general, we demonstrate, on the contrary, that DP can be gained as a perk even without employing any AN. Specifically, we derive a novel bound on DP that converges under general bounded-domain assumptions on model parameters, along with a convergence bound with general smooth and non-convex loss functions. Next, we optimize over receive beamforming and power allocations to characterize the optimal convergence-privacy trade-offs, which also reveal explicit conditions in which DP is achievable without compromising training. Finally, our theoretical findings are validated by extensive numerical results.

Post-hoc calibration of pre-trained models is critical for ensuring reliable inference, especially in safety-critical domains such as healthcare. Conformal Prediction (CP) offers a robust post-hoc calibration framework, providing distribution-free statistical coverage guarantees for prediction sets by leveraging held-out datasets. In this work, we address a decentralized setting where each device has limited calibration data and can communicate only with its neighbors over an arbitrary graph topology. We propose two message-passing-based approaches for achieving reliable inference via CP: quantile-based distributed conformal prediction (Q-DCP) and histogram-based distributed conformal prediction (H-DCP). Q-DCP employs distributed quantile regression enhanced with tailored smoothing and regularization terms to accelerate convergence, while H-DCP uses a consensus-based histogram estimation approach. Through extensive experiments, we investigate the trade-offs between hyperparameter tuning requirements, communication overhead, coverage guarantees, and prediction set sizes across different network topologies. The code of our work is released on: https://github.com/HaifengWen/Distributed-Conformal-Prediction.

Over-the-air federated learning (FL), i.e., AirFL, leverages computing primitively over multiple access channels. A long-standing challenge in AirFL is to achieve coherent signal alignment without relying on expensive channel estimation and feedback. This paper proposes NCAirFL, a CSI-free AirFL scheme based on unbiased non-coherent detection at the edge server. By exploiting binary dithering and a long-term memory based error-compensation mechanism, NCAirFL achieves a convergence rate of order $\mathcal{O}(1/\sqrt{T})$ in terms of the average square norm of the gradient for general non-convex and smooth objectives, where $T$ is the number of communication rounds. Experiments demonstrate the competitive performance of NCAirFL compared to vanilla FL with ideal communications and to coherent transmission-based benchmarks.

Addressing the communication bottleneck inherent in federated learning (FL), over-the-air FL (AirFL) has emerged as a promising solution, which is, however, hampered by deep fading conditions. In this paper, we propose AirFL-Mem, a novel scheme designed to mitigate the impact of deep fading by implementing a \emph{long-term} memory mechanism. Convergence bounds are provided that account for long-term memory, as well as for existing AirFL variants with short-term memory, for general non-convex objectives. The theory demonstrates that AirFL-Mem exhibits the same convergence rate of federated averaging (FedAvg) with ideal communication, while the performance of existing schemes is generally limited by error floors. The theoretical results are also leveraged to propose a novel convex optimization strategy for the truncation threshold used for power control in the presence of Rayleigh fading channels. Experimental results validate the analysis, confirming the advantages of a long-term memory mechanism for the mitigation of deep fading.

This article highlights the potential of a prominent generative artificial intelligence (GAI) method, namely diffusion model (DM), for mobile communications. First, we propose a DM-driven communication architecture which introduces two key paradigms, that is, conditional DM, and DM-driven deep reinforcement learning (DRL), for wireless data generation and communication management, respectively. Then, we discuss the key advantages of the DM-driven communication paradigms. To elaborate further, we explore DM-driven channel generation mechanisms for channel estimation, extrapolation, and feedback in multiple-input multiple-output (MIMO) systems. We showcase the numerical performance of conditional DM using the accurate DeepMIMO channel datasets, revealing its superiority in generating high-fidelity channels and mitigating unforeseen distribution shifts in sophisticated scenes. Furthermore, several DM-driven communication management designs promising to deal with imperfect channels and task-oriented communications are conceived. To inspire future research developments, we highlight the potential applications and open research challenges of DM-driven communications. Code is available at https://github.com/ xiaoxiaxusummer/GAI_COMM/.

With the advent of emerging IoT applications, such as autonomous driving, digital-twin, metaverse, etc., featuring massive data sensing, analyzing, inference, and critical latency in beyond 5G (B5G) networks, edge artificial intelligence (AI) has been proposed to provide high-performance computation of a conventional cloud down to the network edge. Recently, the convergence of wireless sensing, computation, and communication (SC 2 ) for specific edge AI tasks, has aroused a paradigm shift by enabling (partial) sharing of the radio-frequency (RF) transceivers and information processing pipelines among these three fundamental functionalities of IoT. However, most existing design frameworks separate these designs incurring unnecessary signaling overhead and waste of energy, and it is therefore of paramount importance to advance fully integrated sensing, computation, and communication (ISCC) to achieve ultra-reliable and low-latency edge intelligence acquisition. In this article, we provide an overview of principles of enabling ISCC technologies followed by two concrete use cases of edge AI tasks that demonstrate the advantage of task-oriented ISCC, and point out some practical challenges in edge AI design with advanced ISCC solutions.

Design of mechanisms towards deploying over-the-air federated learning (FL) encounters many key challenges. One of them is to comply with the power and bandwidth constraints of the shared channel, while causing minimum deterioration to the learning performance compared to baseline noiseless implementations. For additive white Gaussian noise (AWGN) channels with instantaneous per-device power constraints, prior work has demonstrated the optimality of a power control mechanism based on norm clipping. This was done through the minimization of an upper bound on the optimality gap for smooth learning objectives satisfying the Polyak-Lojasiewicz (PL) condition. In this paper, we make two contributions to the development of AirFL based on norm clipping, which we refer to as AirFL-Clip. First, we provide a convergence bound for AirFL-Clip that applies to general smooth and non-convex learning objectives. Unlike existing results, the derived bound is free from run-specific parameters, thus supporting an offline evaluation. Second, we extend AirFL-Clip to include Top-k sparsification and linear compression. For this generalized protocol, referred to as AirFL-Clip-Comp, we derive a convergence bound for general smooth and non-convex learning objectives. We argue and demonstrate via experiments, that the only time-varying quantities present in the bound can be efficiently estimated offline by leveraging the well-studied properties of sparse recovery algorithms.

This letter investigates a cache-enabled multiuser mobile edge computing (MEC) system with dynamic task arrivals, taking into account the impact of proactive cache placement on the system's overall energy consumption. We consider that an access point (AP) schedules a wireless device (WD) to offload computational tasks while executing the tasks of a finite library in the task caching phase, such that the nearby WDs with the same task request arriving later can directly download the task results in the task arrival and execution phase. We aim for minimizing the system's weighted-sum energy over a finite-time horizon, by jointly optimizing the task caching decision and the MEC execution of the AP, and local computing as well as task offloading of the WDs at each time slot, subject to caching capacity, task causality, and completion deadline constraints. The formulated design problem is a mixed-integer nonlinear program. Under the assumption of fully predicable task arrivals, we first propose a branch-and-bound (BnB) based method to obtain the optimal offline solution. Next, we propose two low-complexity schemes based on convex relaxation and task-popularity, respectively. Finally, numerical results show the benefit of the proposed schemes over existing benchmark schemes.

One of the emerging computing paradigms, mobile edge computing (MEC, also known as fog computing), has been developed to reduce both energy consumption and computation latency for computation-extensive IoT applications. Further, thanks to advantages brought by non-orthogonal multiple access (NOMA) in increasing the capacity of multiple-access channels (MAC), and by service caching in alleviating the burden of responding to repeated computation requests, this paper considers the joint design of communication, computation, and caching for multi-user MEC systems. Aiming for minimizing the weighted-sum energy consumption of communication and computation, given a finite set of computation services, we jointly optimize the NOMA transmission, the computation resources, and the Boolean-variable modeled cache placement, subject to the computation and caching capacity of the edge server as well as the computation latency constraints. To solve the formulated mixed-integer non-convex problem, first, given the cache placement, we solve the non-differentiable convex problem by Lagrangian dual method leveraging a semi-closed form of NOMA transmission power, followed by a one-dimension search for the optimal common task offloading time. Next, an optimal branch-and-bound (BnB) based caching strategy is proposed. Meanwhile, we also provide a heuristic suboptimal cache placement design to reduce computational complexity. Finally, numerical results show the striking performance of the proposed joint optimization of NOMA-based task offloading and service caching compared to the greedy cache placement and other benchmarks without either NOMA-based task offloading or service caching.

With the growing demand for latency-critical and computation-intensive Internet of Things (IoT) services, the IoT-oriented network architecture, mobile-edge computing (MEC), has emerged as a promising technique to reinforce the computation capability of the resource-constrained IoT devices. To exploit the cloud-like functions at the network edge, service caching has been implemented to reuse the computation task input/output data, thus effectively reducing the delay incurred by data retransmissions and repeated execution of the same task. In a multiuser cache-assisted MEC system, users’ preferences for different types of services, possibly dependent on their locations, play an important role in the joint design of communication, computation, and service caching. In this article, we consider multiple representative locations, where users at the same location share the same preference profile for a given set of services. Specifically, by exploiting the location-aware users’ preference profiles, we propose joint optimization of the binary cache placement, the edge computation resource, and the bandwidth (BW) allocation to minimize the expected sum-energy consumption, subject to the BW and the computation limitations as well as the service latency constraints. To effectively solve the mixed-integer nonconvex problem, we propose a deep learning (DL)-based offline cache placement scheme using a novel stochastic quantization-based discrete-action generation method. The proposed hybrid learning framework advocates both benefits from the model-free DL approach and the model-based optimization. The simulations verify that the proposed DL-based scheme saves roughly 33% and 6.69% of energy consumption compared with the greedy caching and the popular caching, respectively, while achieving up to 99.01% of the optimal performance.

To improve the radio spectral efficiency for 5G and beyond, novel radio access techniques need to be designed to accommodate unprecedented number of connected devices, and nonorthogonal multiple access (NOMA) has become a promising candidate. Additionally, power allocation and NOMA-secondary user (SU) assignment technology is an efficient way to enhance the resource utilization efficiency at the power domain and the spectral domain for underlaying cognitive NOMA networks. In this article, first, a joint power allocation and SU assignment problem is formulated for the NOMA downlink transmission in an underlaying cognitive radio network. The worst-case achievable rate for the NOMA-SU is maximized. To solve this mixed-integer nonlinear programming problem, we divide the original optimization problem into two subproblems: NOMA-SU assignment and power allocation. Next, a heuristic algorithm is adopted to solve the NOMA-SU assignment subproblem, and successive convex approximation based method is utilized to design a suboptimal power allocation algorithm. Furthermore, an alternative joint NOMA-SU assignment and power allocation scheme are proposed with its average computational complexity analysis given. Finally, numerical results show that the total throughput for the proposed algorithm outperforms more than 30% compared with an existing benchmark scheme at least.

The proliferation of Internet-of-Things (IoT) devices and cloud-computing applications over siloed data centers is motivating renewed interest in the collaborative training of a shared model by multiple individual clients via federated learning (FL). To improve the communication efficiency of FL implementations in wireless systems, recent works have proposed compression and dimension reduction mechanisms, along with digital and analog transmission schemes that account for channel noise, fading, and interference. The prior art has mainly focused on star topologies consisting of distributed clients and a central server. In contrast, this paper studies FL over wireless device-to-device (D2D) networks by providing theoretical insights into the performance of digital and analog implementations of decentralized stochastic gradient descent (DSGD). First, we introduce generic digital and analog wireless implementations of communication-efficient DSGD algorithms, leveraging random linear coding (RLC) for compression and over-the-air computation (AirComp) for simultaneous analog transmissions. Next, under the assumptions of convexity and connectivity, we provide convergence bounds for both implementations. The results demonstrate the dependence of the optimality gap on the connectivity and on the signal-to-noise ratio (SNR) levels in the network. The analysis is corroborated by experiments on an image-classification task.