Traditional federated learning (FL) approaches face significant challenges when applied to dynamic and heterogeneous Internet of Vehicles (IoV) environments, which are characterized by frequent node mobility, unstable communication links, and highly non-independent and identically distributed (Non-IID) data. In particular, decentralized network topologies exacerbate the difficulty of maintaining model consistency, thereby impairing overall learning performance. To address these challenges, we propose a new hierarchical decentralized federated learning (HDFL) framework. This framework combines the advantages of centralization and decentralization, builds a three-layer collaborative structure, and improves communication flexibility through an asynchronous model exchange mechanism between the edge and the client. Simultaneously, HDFL introduces a local fine-tuning strategy based on knowledge distillation to enhance the generalization ability and stability of the model. Experimental results using an urban traffic simulation platform show that HDFL consistently outperforms representative decentralized FL methods in terms of the achieved accuracy and convergence speed under heterogeneous IoV environments.

This paper introduces a real-time semantic communication system for remote driving, addressing the challenges of video transmission over constrained and fluctuating wireless communication links. Instead of transmitting raw video streams, the proposed system extracts and transmits compact semantic representations, significantly reducing bandwidth requirements while preserving task-relevant visual information. An end-to end semantic encoding-decoding pipeline enables low latency operator-view reconstruction with low latency, improving robustness without relying on high-throughput links. Implemented and evaluated on the real-time prototype, the proposed system demonstrates the practicality of semantic communication for enhancing responsiveness and reliability in remote driving scenarios.

Low-Altitude Economy Networks (LAENets) are emerging as a promising paradigm to support various low-altitude services through integrated air-ground infrastructure. To satisfy low-latency and high-computation demands, the integration of Unmanned Aerial Vehicles (UAVs) with Mobile Edge Computing (MEC) systems plays a vital role, which offloads computing tasks from terminal devices to nearby UAVs, enabling flexible and resilient service provisions for ground users. To promote the development of LAENets, it is significant to achieve low-carbon multi-UAV-assisted MEC networks. However, several challenges hinder this implementation, including the complexity of multi-dimensional UAV modeling and the difficulty of multi-objective coupled optimization. To this end, this paper proposes a novel Retrieval Augmented Generation (RAG)-based Large Language Model (LLM) agent framework for model formulation. Specifically, we develop HybridRAG by combining KeywordRAG, VectorRAG, and GraphRAG, empowering LLM agents to efficiently retrieve structural information from expert databases and generate more accurate optimization problems compared with traditional RAG-based LLM agents. After customizing carbon emission optimization problems for multi-UAV-assisted MEC networks, we propose a Double Regularization Diffusion-enhanced Soft Actor-Critic (R2DSAC) algorithm to solve the formulated multi-objective optimization problem. The R2DSAC algorithm incorporates diffusion entropy regularization and action entropy regularization to improve the performance of the diffusion policy. Furthermore, we dynamically mask unimportant neurons in the actor network to reduce the carbon emissions associated with model training. Simulation results demonstrate the reliability of the proposed HybridRAG-based LLM agent framework, which achieves a$6.6\%$improvement in F1 scores over traditional RAG, and validate the effectiveness of the R2DSAC algorithm, which outperforms the SAC algorithm by up to$64.17\%$.

Mobile metaverses are envisioned as a transformative digital ecosystem that delivers immersive, intelligent, and ubiquitous services through mobile devices. Driven by Large Language Models (LLMs) and Vision-Language Models (VLMs), Artificial Intelligence (AI) agents hold the potential to empower the creation, maintenance, and evolution of mobile metaverses, enabling seamless human-machine interaction and dynamic service adaptation. Currently, AI agents are primarily built upon cloud-based LLMs and VLMs. However, several challenges hinder their efficient deployment, including high service latency and a risk of sensitive data leakage during perception and processing. In this paper, we develop an edge-cloud collaboration-based federated AI agent construction framework in mobile metaverses. Specifically, Edge Servers (ESs), as agent infrastructures, first create agent modules in a distributed manner. The cloud server then integrates these modules into AI agents and deploys them at the edge, thereby enabling low-latency AI agent services for users. Considering that ESs may exhibit dynamic levels of willingness to participate in federated AI agent construction, we design a two-period dynamic contract model to continuously incentivize ESs to participate in agent module creation, effectively addressing the dynamic information asymmetry between the cloud server and ESs. Furthermore, we propose an Enhanced Diffusion Model-based Soft Actor-Critic (EDMSAC) algorithm to effectively generate optimal dynamic contracts. In the algorithm, we apply dynamic structured pruning to DM-based actor networks to enhance denoising efficiency and policy learning performance. Simulation results demonstrate that the EDMSAC algorithm outperforms the DMSAC algorithm by up to 23% in optimal dynamic contract generation.

Modern GPU clusters increasingly comprise diverse heterogeneous GPUs, driven by the continuous release of new GPU models. Achieving a balance between fairness and efficiency when scheduling multi-tenant Deep Learning (DL) training jobs on such clusters is inherently challenging. Existing DL training schedulers largely emphasize fairness through GPU temporal sharing, while the spatial dimension of resource allocation is often underexplored. This oversight can lead to GPU fragmentation and suboptimal system performance. In this paper, we propose STS-Fairness, a spatiotemporal sharing fairness scheduler. STS-Fairness partitions each GPU into multiple isolated slots under a novel spatiotemporal fairness constraint and allocates jobs using a round-based allocation mechanism. We guarantee that STS-Fairness achieves overall performance optimality while satisfying spatiotemporal fairness constraints. The scheduling problem is formulated as an integer nonlinear program (INLP) that is solved to optimality in polynomial time via dynamic programming. We deployed the STS-Fairness framework on both physical and simulated heterogeneous clusters and conducted large-scale experiments. These results demonstrate that STS-Fairness reduces average JCT by$1.2 \times$, shortens makespan by$1.24 \times$, and increases throughput by$\mathbf{1. 2 5} \times$compared to state-of-the-art (SoTA) schedulers.

Artificial intelligence-generated content (AIGC) is increasingly featuring a key to extract intent information from external instructions and generate required content in digital twin (DT)-enabled application scenarios. To construct DT contexts as the input of generative artificial intelligence (GenAI) algorithms, space–air–ground integrated networks (SAGINs) with hierarchical structures can facilitate object cloning from the physical world to a virtual space within vast geographical regions. In this work, we propose a novel DT synchronization framework residing in SAGINs to provide AIGC services. Autonomous aerial vehicles (AAVs) are in charge of gathering real-time information from the external environment and transmitting synchronization data to the core cloud via a communication relay, i.e., base station (BS) or satellite. In the proposed framework, we develop a resource allocation problem for DT synchronization, aiming to minimize the time-average energy costs of AAVs under the constraints on resource provision and long-term transmission queue stability. To address the complexity and dynamics of SAGINs, we first transform the original resource allocation problem into several deterministic problems based on the Lyapunov optimization. Then, a diffusion model-based resource allocation (DRA) algorithm is developed to solve the deterministic problem in each time slot, where a novel diffusion model is proposed to generate integer relay selection decisions with the aid of auxiliary gradients provided by conventional model-based optimization. Finally, we provide theoretical and simulation evaluations to demonstrate that the DRA algorithm can reduce energy consumption and improve resource utilization by comparing it with deep reinforcement learning (DRL) and heuristic algorithms.

The widespread application of AI-generated content (AIGC) services has driven demand for efficient computational resources, making effective task scheduling and computation offloading in edge computing (EC) environments a critical research topic. However, the high computational requirements and low latency demands of AIGC services, combined with the limitations of EC, present challenges for existing offloading methods, such as unstable decision making in dynamic task environments and resource overloading. Here, we propose a decentralized AIGC task offloading architecture within an IoT-assisted EC network to optimize the quality of AIGC services. In this architecture, we define a multiobjective joint optimization problem for AIGC task offloading, aiming to simultaneously optimize key performance metrics, such as task latency, energy efficiency, and load balancing. To address this problem, we introduce an improved proximal policy optimization (PPO)-based deep reinforcement learning (DRL) algorithm, named TOPPO. By incorporating a policy update step size constraint and a clipping mechanism, TOPPO significantly enhances the stability of the training process and reduces fluctuations during policy updates. Additionally, the algorithm integrates an LSTM model to improve its ability to handle temporal dependencies. Through continuous interaction between the model and the environment, the offloading strategy is iteratively updated to ensure that diverse AIGC tasks are efficiently executed on IoT devices or edge servers. Extensive simulations and performance evaluations demonstrate that the proposed method achieves significant improvements in task latency, energy consumption, and load management during AIGC task processing.

The distribution of the labeled data can greatly affect the performance of a semi-supervised learning (SSL) model. Most existing SSL models select the labeled data randomly and equally allocate the labeling quota among the classes, leading to considerable unstableness and degeneration of performance. This study unsupervisedly constructs a leading forest that forms another metric space, based on which it is convenient to define the fuzzy membership function to characterize central and divergent samples and select both types with fuzzy Xor logic. The labeling quota can thus be allocated adaptively among different classes. The proposed determinate labeling strategy can generally improve the performance for most SSLs. Especially, when combined with the kernelized large margin component analysis, it produces a novel semi-supervised classification model. In addition, the multi-modal issue in SSL is effectively addressed by the multi-granular structure of leading forest that readily facilitates multiple local metrics learning. Extensive experimental results demonstrate that the proposed method achieved competitive efficiency and encouraging accuracy when compared with the state-of-the-art methods

With the enhancement of health awareness and the development of artificial intelligent technology, healthcare recommender systems (HRS) play an increasingly important role in individual health management. Meanwhile, the widespread usage of large models has significantly improved the efficiency and accuracy in the analysis and utilization of medical data. In this paper, we comprehensively summarize the basic types of recommender systems as well as the new trends in utilizing large models. Then we introduce the recommendation applications in healthcare areas from six aspects, i.e., disease risk prediction, medication recommendation, medical resource recommendation, mental health support, health life management, and health education. At last, we explore some current issues and challenges within HRS, as well as the development of potential solutions and directions in the future.

Metaverse is an immersive, seamless, interactive, comprehensive virtual world, as well as a replication, extension, and transcendence of the real world. Unmanned aerial vehicle (UAV)-enabled mobile edge computing (MEC) is becoming a key technology for ubiquitous Metaverse services. To enhance network resource utilization, we introduce dual connectivity (DC) technologies in UAV-enabled MEC, which increases the time complexity associated with resource management. Considering the specific features of DC communication channels, we propose a UAV-assisted Metaverse rendering problem to enhance the Metaverse service experience and reduce the energy cost of edge devices. To solve the rendering problem with low complexity, we propose a diffusion model-based Metaverse rendering algorithm, where a novel diffusion model is used to generate integer rendering decisions with the aid of the gradient provided by the model-based Metaverse rendering problem. Moreover, with the given rendering decisions, the communication and computation resource allocation results are derived by the model-based optimization method. Finally, we conduct extensive simulation experiments based on real-world datasets. Comprehensive simulation results demonstrate that the diffusion model-based Metaverse rendering algorithm can reduce the Metaverse frame rendering time and improve user experience.

The advanced sixth-generation (6G) wireless network is considered as an indispensable part of the Metaverse, where a substantial volume of communication content is transmitted through multiple modalities, placing significant transmission loads on communication channels. In this paper, we propose a framework for multi-modal semantic communication using hashing-based semantic extraction approach to produce optimal binary signatures (hash codes). Instead of directly using coarse-grained feature fusion methods, we capture deep semantics in self-attention manner, achieving fine-grained multi-modal feature fusion thereby strengthening the representation ability of hash codes. To enhance adaptability in practical situations, we then design a modality-completion module to address missing modalities in data, accommodating scenarios with both single-modal and cross-modal data. We evaluate the proposed semantic extraction framework on two popular multi-modal datasets, comparing it with the latest hashing methods and then demonstrate the effectiveness in various channel conditions.

Semantic communication is an emerging paradigm for digital twin (DT) synchronization in unmanned aerial vehicle (UAV)-assisted edge computing environments, where machine learning (ML) models are deployed on edge servers and UAVs as semantic encoders and decoders to perform real-time synchronization. However, with limited system resources, additional computation workloads are still brought to all participants for semantic information extraction and recovery. In this work, we propose an optimized tiny ML-based DT synchronization framework to minimize the synchronization latency in UAV-assisted edge computing environments, considering time-average constraints on virtual energy deficit queue stability. Due to the coexistence of tiny ML-based semantic communications, a semantic extraction factor is introduced to formulate the DT synchronization problem as a time-average time minimization problem. By leveraging the Lyapunov optimization framework, the multi-stage DT synchronization problem is transformed into several per-slot resource allocation problems. To solve the per-slot optimization problem efficiently, a deep reinforcement learning-based synchronization (DRLS) algorithm is proposed, where an actor-critic structure is adopted to generate synchronization actions with low time complexity. Finally, we conduct simulation experiments to evaluate the performance of the proposed DRLS scheme. Numerical results demonstrate that our DRLS algorithm can reduce 8.23% of DT synchronization delay and 15.31% of synchronization data dropping rates on average by comparing it with the UAV-edge collaborative synchronization scheme without semantic communications. Besides, the DRLS algorithm can achieve up to 57.14% synchronization energy reduction compared with representative synchronization policies.

Unmanned aerial vehicle (UAV)-assisted mobile edge computing (MEC) has largely extended the border and capacity of artificial intelligence of things (AIoT) by providing a key element for enabling flexible distributed data inputs, computing capacity, and high mobility. To enhance data privacy for AIoT applications, federated learning (FL) is becoming a potential solution to perform training tasks locally on distributed IoT devices. However, with the limited onboard resources and battery capacity of each UAV node, optimization is required to achieve a large-scale and high-precision FL scheme. In this work, an optimized multi-UAV-assisted FL framework is designed, where regular IoT devices are in charge of performing training tasks, and multiple UAVs are leveraged to execute local and global aggregation tasks. An online resource allocation (ORA) algorithm is proposed to minimize the training latency by jointly deciding the selection decisions of clients and a global aggregation server. By leveraging the Lyapunov optimization technique, virtual energy queues are studied to depict the energy deficit. With the help of the actor-critic learning framework, a deep reinforcement learning (DRL) scheme is designed to improve per-round training performance. A deep neural network (DNN)-based actor module is designed to derive client selection decisions, and a critic module is proposed through a conventional optimization method to evaluate the obtained selection decisions. Moreover, a greedy scheme is developed to find the optimal global aggregation server. Finally, extensive simulation results demonstrate that the proposed ORA algorithm can achieve optimal training latency and energy consumption under various system settings.

Metaverse is envisaged as an evolving Internet paradigm that allows people to play, work, and socialize in a shared and virtual ecosystem with immersive and seamless experiences. However, multiple users will access the metaverse world for diverse scenes simultaneously due to its social property. How to provide high-quality and low-latency metaverse services for massive concurrent users is a crucial problem. In this work, a novel social-aware edge caching (SEC) framework is proposed for metaverse systems, where metaverse scenes are divided into massive environment panoramic frames and dynamic objects with different priorities. An unmanned aerial vehicle (UAV)-assisted edge server is deployed to cache the environment panoramic frames, while the dynamic objects are rendered on head-mounted displays (HM Ds). A synchronous advantage actor-critic (SA2C) algorithm is developed to generate caching solutions with low time complexity by considering the collective behaviors and social dynamics for requesting similar scenes. Finally, we provide some simulation experiments by leveraging a real-world dataset. The numerical results reveal that the proposed algorithm can reduce the service time and increase the cache hit rate significantly by comparing it with two benchmark caching algorithms.

The collaborative edge and cloud computing system has emerged as a promising solution to fulfill the unprecedented high requirements of 5G application scenarios. Due to vendor variations, it is often difficult to manage hardware facilities in such a collaborative system. Moreover, the amount of data generated and tasks requested by end devices are increasing exponentially, which introduces storage and computation bottlenecks. To address these issues, a novel systematic framework called software-defined edge and cloud computing (SD-ECC) is designed to manage the underlying physical resources of edge and cloud layers via software. SD-ECC is combined with an erasure-coded storage system, for which a task scheduling problem is formulated by considering data access and task processing steps. Then, a joint data access and task processing (JDATP) algorithm is proposed to minimize the task response time including data access latency and task processing latency. A practical SD-ECC platform is developed on OpenStack, OpenDaylight, and Kubernetes to conduct experiments with real-world datasets. The experimental results demonstrate that our proposed JDATP algorithm can reduce 20.87% of the task response time and increase 14.16% of the remaining storage space on average by comparing it with alternative schemes.

The edge-cloud computing and network slicing have emerged as promising solutions to fulfill the diversity of IoT applications enabled by 5G and beyond. However, edge-cloud computing systems are composed of various hardware facilities, leading to difficulties in hardware control and management. With network slicing, underlying resource sharing among multiple slice users is allowed, leading to potential attacks to the slice formulation processes and malicious usage of network slices that may result in inefficient resource utilization of the system. To address the aforementioned network slice security issue, we first propose a new systematic framework, named software-defined edge-cloud computing (SD-ECC), which applies standard software to control the hardware infrastructure regardless of vendor variations. With SD-ECC, resource slices are formulated by including storage and computational resources provided by edge and cloud servers. Then, we study an optimal slicing-based resource orchestration problem by considering slice-initiated attacks as possible adversaries, which includes both interslice and intraslice resource orchestrations. A secure slicing-based resource orchestration (SS-RO) algorithm is designed by minimizing the delay and resource utilization simultaneously to mitigate the impacts of the slice-initiated attacks, where the Benders decomposition is employed to obtain the interslice orchestration outcome, and a quadratic transformation method is applied to derive the intraslice orchestration solution. The experimental results demonstrate that the proposed SS-RO algorithm outperforms baseline schemes in terms of the ratio of accepted attacking tasks, energy consumption, and system throughput.