Unmanned Aerial Vehicles (UAVs) are widely used in Mobile Edge Computing (MEC) and provide computing power at the network edge due to their flexible deployment, reliable wireless communication, and wide coverage. However, coordinating and optimising the relationship between UAV deployment, task offloading, and resource allocation poses a great challenge due to the tight interdependence between the three. In this paper, we study the problem of task offloading and computational resource allocation in a UAV-assisted MEC system with multiple UAVs collaborating, highlighting UAV deployment strategies. We define the system average user cost based on system energy consumption and latency. In this paper, the joint optimisation of UAV deployment, task offloading decisions, and computational resource allocation is formulated as a mixed integer Nonlinear programming (MINLP) problem with the objective of minimising the system average user cost. To address the combinatorial complexity problem, we develop a two-stage optimisation approach: 1) a Joint Linear Programming and Greedy Policy Algorithm (JLPGA) for efficient UAV deployment; 2) a Computation Offloading Decision and Resource Allocation Algorithm Based on Twin Delayed Deep Deterministic Policy Gradient (CODRA-TD3) for dynamic task offloading and resource allocation. Evaluation results show that our proposed algorithm outperforms the other three baseline algorithms in minimising the average user cost of the system.

Leveraging wireless caching in unmanned aerial vehicles (UAVs)-assisted edge networks, UAVs can pre-cache part of popular contents and fly to the users directly, the repeated backhaul transmission delays are reduced. However, the limited caching space, complex network environments and varying user demands make a challenge for designing an effective caching content selection and update scheme. The caching update of UAVs introduces an additional cost, and UAVs can move to the suitable locations in real time to improve the air-to-ground transmission rates. In this paper, we propose a Variable Dual time scale Caching Update and Resource Allocation scheme (VD-CURA) for UAV-assisted edge networks, to minimize the average system task processing latency per unit content size. A joint optimization problem is proposed that the caching update for UAVs, including the caching content selection and update interval are optimized at the large time scale, and the UAV trajectory optimization and bandwidth allocation are performed at each small scale. For the varying networks, a hierarchical reinforcement learning (HRL)-based VD-CURA framework with two layers is proposed, with combined proximal policy optimization (PPO) in the upper layer and deep deterministic policy gradient (DDPG) in the lower layer. Notably, we introduce a new caching content selection metric Experimental results demonstrate the robust convergence and adaptability of the proposed VD-CURA solution algorithm, compared with the baseline schemes, the average latency per unit of content is reduced significantly.

Mobile communication is a cornerstone technology in our society, facilitating personal relationships, fostering cooperation, and advancing education. Each generation of this technology has brought new services, from voice calls and messaging to internet access, and now Artificial Intelligence (AI). This transformation has steadily augmented the mobile network resource requirements. Driven by the societal and economical needs to increase the mobile network energy efficiency, reduce its energy consumption and the associated environmental footprint, this comment surveys standardized metrics to assess energy-related network performance and overview data-driven state-of-the-art solutions to accelerate the energy transition of mobile networks.

Unmanned Aerial Vehicles (UAVs) can assist traffic management in performing collaborative inference to address vehicle related demands in congested scenarios, such as minor accident identification. However, constrained by UAV computational heterogeneity, insufficient communication bandwidth and limited cache capacity, unstructured multi-UAV assistance schemes yield suboptimal performance. Efficient and low-latency distributed inference in dynamic environments remains an urgent challenge in Internet of Vehicles (IoVs). In this paper, we propose a multi-UAV collaborative inference scheme based on dynamic partitioning and cache optimization, establishing a systematic framework that encompasses system architecture, UAV modeling, task and inference model design, caching mechanisms, and latency modeling. Then, a joint optimization problem is formulated and decomposed into three subproblems: dynamic model partitioning, cache scheduling, and UAV position optimization. A decision framework leveraging the Proximal Policy Optimization (PPO) algorithm is designed to realize adaptive optimization for dynamic environments, ensuring near-optimal performance while maintaining low decision-making overhead. Extensive simulation results demonstrate that the proposed optimization scheme outperforms existing baselines in terms of average inference latency, task completion rate, and cache efficiency. Notably, it exhibits superior adaptability and robustness in scenarios with high task loads and significant channel condition fluctuations.

In the Space-Air-Ground Integrated Network (SAGIN), ground users exhibit high mobility and randomly generate heterogeneous computing tasks across different times and regions, posing significant challenges to dynamic perception and intelligent decision-making in Mobile Edge Computing (MEC) for task offloading and resource allocation. Given the limited computational resources of Unmanned Aerial Vehicles (UAVs) and satellite nodes, accurately identifying future congestion-prone areas and efficiently scheduling resources to minimize system cost becomes a key issue. To address this, this paper proposes an offloading optimization approach that integrates dynamic region partitioning, congestion prediction, and deep reinforcement learning (DRL). Specifically, an improved K-means algorithm is employed to dynamically cluster ground users, with each UAV acting as the cluster center to provide edge computing services. Considering the continuous mobility of users and highly variable task loads, a Long Short-Term Memory (LSTM) network is introduced to predict future congestion levels, which are then incorporated into the state representation to enhance the perception capability of the DRL policy. The Proximal Policy Optimization (PPO) algorithm is subsequently used to jointly optimize task offloading decisions and resource allocation strategies. The simulation results demonstrate that the proposed method effectively reduces the average user cost in the MEC system compared to three baseline algorithms.

With the continuous growth of communication demands, 6G Space-air-ground integrated Network (SAGIN) has demonstrated significant value in mobile edge computing (MEC) in low-coverage areas such as disaster-stricken regions, mountainous regions, and the ocean. However, limited by the high dynamics and multi-layer heterogeneity of SAGIN, designing efficient task offloading and resource scheduling strategies is crucial to improve the Quality of Service (QoS) of the network. To address the challenges posed by user mobility, UAV trajectory variation, and task heterogeneity, we construct a joint optimization problem that integrates task offloading, UAV trajectory planning, and resource scheduling, aiming to minimize the system processing delay. To solve the problem, this paper proposes a Mobility-aware Graph Neural Network (GNN)-Enhanced Deep Reinforcement Learning (DRL) (MGN-DRL) algorithm. Firstly, GNN are utilized to mine the spatial topological relationships among mobile users (MUs) and predict their future trajectories, providing future state information for the DRL model. Subsequently, the DRL integrates the predicted state with the current environmental perception to jointly optimize multi-dimensional decision variables. Finally, the simulation experiments show that the proposed MGN-DRL joint optimization framework is significantly superior to the baseline algorithm, effectively improving the system QoS and the quality of experience (QoE) of MUs.

Unmanned aerial vehicles (UAVs), with their exceptional flexibility and mobility, can be rapidly deployed to complex regions where traditional terrestrial infrastructure struggles to provide coverage, such as remote areas and disaster-stricken sites. It effectively addresses the challenges of limited network coverage and high deployment costs in conventional terrestrial mobile edge computing (MEC) networks. As latency-sensitive applications such as the Internet of Things (IoT) and autonomous driving become increasingly prevalent, end-to-end delay has emerged as a vital measure for evaluating quality of service (QoS). The joint optimization of task scheduling strategies, offloading ratios, and UAV flight trajectories directly determines the efficiency of data transmission and computational processing. However, existing studies on delay optimization in UAV-assisted MEC generally overlook the constraint of maximum UAV energy consumption or fail to effectively handle complex decision-making in continuous action spaces. To address these challenges, this paper proposes a novel joint optimization approach based on the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm, which collaboratively models task offloading, trajectory planning, and resource allocation under strict energy constraints. This study models the problem using the framework of a Markov Decision Process (MDP). Simulation results demonstrate that the proposed algorithm achieves a reduced average delay compared to baseline algorithms while maintaining stable policy convergence performance. This work provides a viable solution for the low-latency and reliable deployment of UAV-assisted MEC systems in real-world scenarios.

Achieving ubiquitous network coverage has emerged as a fundamental objective in the evolution of 6G wireless communication systems. However, the deployment of communication infrastructure in remote and underserved areas remains prohibitively expensive and traditional base station architectures and edge computing paradigms often fall short in meeting the computational offloading demands in such scenarios, thereby limiting the coverage and service capabilities envisioned for 6G networks. To address these challenges, this paper proposes a computation offloading framework based on an integrated space-air-ground architecture, which enables over-the-air computation and cooperative scheduling capabilities. Specifically, a two-level scheduling mechanism is designed: a ground-level task aggregation strategy based on multi-hop Ant Colony Optimization (ACO), and an aerial-level intelligent caching algorithm employing a multi-agent Deep Deterministic Policy Gradient (MADDPG) approach. The latter is further enhanced by a heuristic pretraining phase to alleviate the cold-start problem and improve convergence efficiency. Extensive simulation results demonstrate that the proposed approach significantly outperforms existing baseline schemes in terms of key performance metrics, including average task delay and completion ratio

Edge computing and integrated sensing and communication (ISAC) technologies offer promising prospects for intelligent transportation systems (ITSs) in which the sensing data of vehicles can be processed directly or be offloaded to a base station (BS) or to the surrounding vehicles. However, the inherent scarcity of communication resources becomes a crucial problem in ITSs, especially when ISAC is introduced. In this paper, we propose an ISAC-assisted vehicular edge computing networks (VECNs) architecture composed of two interconnected stages: resource management and task offloading. Vehicles perform sensing and dynamically offload sensing tasks to the BS or nearby vehicles based on the link conditions. A two-stage joint optimization problem is formulated to optimize the resource block (RB) allocation for V2I and V2V links, including communication and sensing power among multiple vehicles, so as to maximize the overall data transmission rate. Concurrently, the offloading decisions are optimized, aiming to minimize the weighted sum of the system task completion delay and energy consumption. Considering the complex, dynamic transmission environment, we reformulate these problems as Markov Decision Processes and propose a deep reinforcement learning-based dual-stage resource management and offloading decision strategy (DDROS). Simulation results demonstrate that the proposed DDROS achieves strong convergence and exhibits significant performance advantages over baseline strategies under various conditions.

Following the trend of electricity marketization, the power distribution system needs to conduct accurate load forecasting to improve regional power distribution strategies. Meanwhile, power users emphasize the protection of electricity data privacy. It is challenging to simultaneously meet the requests for accurate regional-level load forecasting for distribution systems and fine-grained user-level load forecasting for power users. In this paper, we develop a multi-level short-term load forecasting (STLF) model based on an improved federated learning (FL) framework aimed at serving both distribution systems and power users. To construct a local load prediction model, the weather and historical load data are utilized, leveraging the strong correlation between short-term power load data and long sequence. A feature attention mechanism-enhanced bidirectional superimposed recurrent neural network (Fam-Bi-SRNN) is proposed. The federated average algorithm (FedAvg) and adaptive sample weighting (ASW) strategy are combined to build an improved FL prediction framework that integrates regional-level and user-level weight adjustments. Based on simulation examples using 96 real-time load datasets, the model has achieved a significant improvement in communication efficiency. Regional-level and user-level root mean square errors (RMSE) reach 0.025 and 0.018, outperforming the traditional algorithms. Simulation results show the model’s ability to ensure rapid training, high accuracy, and multi-level prediction while protecting user data privacy.

With the growing number of mobile devices (MDs), users can achieve higher quality of service (QoS) by offloading latency-sensitive tasks to edge access points (E-APs). However, the number of MDs that E-APs can serve is limited. Consequently, the existing network cannot fully meet users’ real-time demands. In recent years, advancements in terahertz (THz) band technology, with its ultra-high data transmission rate and bandwidth, have opened new possibilities for addressing this issue. This paper explores a unmanned aerial vehicle (UAV) assisted computation offloading network utilizing THz technology. The study examines computation offloading and resource allocation under constraints, including energy consumption and resource availability. A Deep Deterministic Policy Gradient (DDPG) based resource allocation and offloading and UAV Placement Algorithm (DRAOP) is proposed to find the optimal solution in a stochastic dynamic environment. Simulation results demonstrate the effectiveness of the proposed scheme.

This letter proposes a matching scheme within non-orthogonal multiple access-integrated sensing and communication (NOMA-ISAC) based uncrewed aerial vehicle (UAV) networks. NOMA-ISAC technology is utilized to optimize the downlink transmission delay (TD) and conditional mutual information (MI) performance for UAVs. The user clustering, flight trajectory and power allocation of UAVs are optimized jointly. Recognizing the varying population density of users, an improved density-weighted agglomerative hierarchical clustering (DWAHC) algorithm and a dynamic load balancing algorithm are applied to assign each user to the appropriate UAV. Then, a multi-agent deep reinforcement learning algorithm is employed to optimize the UAVs'trajectory and transmission power through interactive learning. Experimental results demonstrate that the proposed scheme outperforms orthogonal multiple access (OMA) and other baseline algorithms in terms of reducing TD and improving conditional MI.

While originally designed for the cloud, the benefits of the serverless paradigm are vital in Edge/Fog computing environments. In this paper, we propose Ekko, a novel decentralized edge serverless scheduling system, which enables a large number of serverless applications to run simultaneously at the edge through the Functionas-a-Service (FaaS) model. The key insight is to re-architect the common centralized or hierarchical scheduling systems into a fully decentralized one by using the distributed hash table (DHT) based peer-to-peer (P2P) model, in which many distributed schedulers operate autonomously without any centralized state. In sharp contrast to existing studies, any edge node in our system can act as a scheduler, a function worker, a query forwarder, or a storage node, and flexibly switch between these roles, thereby significantly improving scalability and adaptivity. Ekko introduces three design innovations: a boundary-aware P2P organization, distributed shadow schedulers with a keychain scheduling algorithm, and a distributed locality-aware bucket image store. Our evaluation on 500 Amazon EC2 nodes shows that, compared to the state-of-the-art, Ekko reduces the 90-th percentile tail queue wait time by up to 96.6 %, the scheduling time by up to 38.5 %, and the total deployment time by up to 89.5 %, while efficiently scaling to millions of function invocation requests on thousands of edge nodes.

Connected and automated vehicles (CAVs) have emerged as an efficient solution to improve the driving experience in the intelligent transportation systems (ITSs), in which the targeted vehicle (TV) can switch between the human-driven (HD) and autonomous-driven (AD) modes to act as server or terminal in vehicular edge computing networks (VECNs). However, due to the dynamic nature of traffic networks and the moving of vehicles, distribution of computational resources is imbalanced and variable, it is a challenge to design the cooperative resource management scheme for the whole journey of vehicle users. In this article, we propose a joint driving model selection and resource management scheme for TV in each road segment, to maximize the vehicle users' satisfaction of the whole journey. For the complex formulated joint optimization problem, we design a three-stage hierarchical optimization (3SHO) framework, using deep Q-network (DQN) for driving mode optimization in the first stage and deep deterministic policy gradient (DDPG) for optimizing resource management under different selected driving modes. And a terminal-server matching mechanism is introduced to enable dynamic service quality improvement for TV. Specially, we design a new user satisfaction function with the quality of service, traffic revenue, and the gap between expected and actual revenues of users are considered. Experimental results showcase the robust convergence of the 3SHO algorithm, the adeptness to dynamic traffic networks, and the capacity to enhance user satisfaction significantly.

The complexity and unpredictability of disaster areas pose significant challenges to post-disaster rescue operations. UAVs, with their flexibility and rapid response capability, are able to perform various tasks and provide critical field data. However, these tasks are often computationally intensive and have strict latency requirements, and UAVs have limited computational resources. Therefore, effectively handling a large number of real-time tasks becomes a significant challenge. To address this, we propose a solution combining UAVs and mobile edge computing to optimize task processing. We focus on UAV-assisted task computation in post-disaster scenarios, considering indivisible and latency-sensitive tasks, and construct an edge computing framework for task offloading. We propose a deep reinforcement learning-based task offloading algorithm (TODRL), which predicts UAV loads using LSTM and adjusts offloading strategies with Dueling DQN to minimize latency. Experimental results show significant latency reduction compared to traditional methods, validating the approach's potential and effectiveness in postdisaster rescue.

The rapid advancement of 5G and 6G technologies has introduced various innovative applications, such as autonomous driving and augmented reality, which significantly increase network traffic and computational demands, particularly in densely populated areas. This study focuses on the utilization of unmanned aerial vehicles (LAN's) for Mobile Edge Computing (MEC) to address these challenges. Specifically, we explore the joint optimization of base station (BS) selection, computing resource allocation and UAV trajectory to minimize system delays and energy consumption in scenarios where computational requirements are related to user movement. A deep reinforcement learning -based approach, UM-DDPG, is proposed to optimize task offloading strategies and UAV trajectories. With our simulation platform, the results show that our method has been effective in reducing the system cost, including both delay and energy consumption, compared to traditional methods.

In the Collaborative Vehicle-Infrastructure System (CVIS), comprehensive awareness of the surrounding road environment is crucial for ensuring safe driving and enhancing the user driving experience. However, due to the high mobility of vehicles, the large volume of task data generated, and the inherent limitations in range and accuracy of onboard sensor coverage, traditional Vehicle Edge Computing (VEC) faces significant challenges in efficient task offloading. Meanwhile, when a Roadside Unit (RSU) receives a large volume of task data transmissions from vehicles, it often results to congestion and excessive overhead. To address these issues, this paper proposes an intelligent partial reverse offloading strategy, namely, the Deep Deterministic Policy Gradient (DDPG) algorithm integrating Spatial-Temporal Affinity Clustering (STAC) and Task Priority Scheduling (SP-DDPG). First, to determine the optimal offloading regions in advance, a clustering algorithm is proposed to reduce the decision space and avoid inefficient offloading decisions. Second, a task priority scheduling algorithm is proposed to prioritize tasks with high computational demand and critically limited residence time within the RSU's coverage area. Finally, Deep reinforcement learning (DRL) techniques are used to generate offloading decisions, jointly optimizing computation latency and vehicle service costs. Simulation results demonstrate that, compared to existing schemes, the proposed algorithm significantly reduces computation latency while maintaining the lowest service cost.

The growing complexity of the sixth-generation (6G) wireless networks has positioned artificial intelligence (AI) as a critical tool for radio resource management (RRM). However, the opacity of AI learning models undermines their trustworthiness, robustness, and interpretability, significantly impeding their application in network resource management for 6G scenarios. Establishing reliable edge intelligence (EI) is crucial for the future of various domains, particularly in virtual reality (VR). By leveraging EI, VR applications can better meet user demands, enhancing service experiences and reliability. The 360-degree video characteristics challenge existing VR transmission networks. We propose a unmanned aerial vehicle (UAV)-assisted joint multicast-unicast system to optimize user QoE by jointly optimizing UAV trajectories, user associations, quality requests, user pairing, and power allocation. To address this optimization problem, we first decompose it into inter-frame and intra-frame components, and propose a twin delayed deep deterministic policy gradient (TD3)-based algorithm to tackle the inter-frame problem. After solving the inter-frame problem, we utilize the quantum particle swarm optimization (QPSO) algorithm and matching theory to address the intra-frame problem. Simulation results demonstrate that our proposed algorithm, significantly improves the QoE for users.

In this paper, we aim to save the total energy consumption of servers through elastic scaling of CPU resources in container cloud. To be practical, we propose an online scheduling method, which consists of three parts: container placement, vertical scaling and migration. 1) For container placement, we design an algorithm based on dynamic threshold, resource balancing and delayed running. When there are PMs (Physical Machines) turned on, the CPU threshold increases so that the containers can be placed onto fewest possible PMs. To make full use of multi-dimensional resources of PM, we put forward a resource balancing strategy. Since the number of CPU cores can be scaled dynamically in containers’ run time, the start time of containers can be delayed without violating deadlines. 2) For vertical scaling, a collaborative multi-agent reinforcement learning (MARL) algorithm is proposed to adjust the container’s CPU, so that the containers on the same PM can finish simultaneously if possible. Then, the PM can be turned off to save energy. 3) To further reduce total energy consumption, we consider migrating the containers from underloaded PMs and overloaded PMs. Experiment results show the superior performance of our method to that of the state-of-the-art.

Nowadays, many companies possess various types of AI accelerators, forming heterogeneous clusters. Efficiently leveraging these clusters for high-throughput large language model (LLM) inference services can significantly reduce costs and expedite task processing. However, LLM inference on heterogeneous clusters presents two main challenges. Firstly, different deployment configurations can result in vastly different performance. The number of possible configurations is large, and evaluating the effectiveness of a specific setup is complex. Thus, finding an optimal configuration is not an easy task. Secondly, LLM inference instances within a heterogeneous cluster possess varying processing capacities, leading to different processing speeds for handling inference requests. Evaluating these capacities and designing a request scheduling algorithm that fully maximizes the potential of each instance is challenging. In this paper, we propose a high-throughput inference service system on heterogeneous clusters. First, the deployment configuration is optimized by modeling the resource amount and expected throughput and using the exhaustive search method. Second, a novel mechanism is proposed to schedule requests among instances, which fully considers the different processing capabilities of various instances. Extensive experiments show that the proposed scheduler improves throughput by 122.5 clusters, respectively.