关键词： Mobile edge computing   Deep reinforcement learning   Mixture-of-expert   Workload cluster   Edge collaboration  

摘要： In the context of efficient collaboration on 5G heterogeneous networks, computation and communication, mobile edge computing (MEC) and cloud-edge collaboration encounter several issues. These include multi-agent cooperation in deep reinforcement learning (DRL) for multi-task processing, the efficiency of action decisions by mobile edge computing server (MECS), and the ineffectiveness of traditional DRL algorithms in resource allocation. To address these challenges, a framework consisting of multiple wireless mobile terminals (WMTs) enabled by unmanned aerial vehicle (UAV) and multiple MEC servers is constructed, considering diversity and priorities of the workload generated by WMTs. Furthermore, to optimize edge collaborative workload offloading, migrating, and resource allocation decisions for minimizing the weighted cost of workload processing, we propose a hybrid DRL approach combining the dueling double deep Q-network (D3QN) and deep deterministic policy gradient (DDPG) algorithms, named OMRA-DRL. Regarding the proposed OMRA-DRL, a K-Means-based clustering algorithm groups similar workloads to simplify optimization. Additionally, a mixture-of-expert (MoE) system enables efficient action selection. Along with D3QN for better MECS selection, the maximum advantage selection strategy of the advantage function (MMSS-D3QN) is formulated to migrate workload clusters to multiple MECSs, achieving multi-edge cooperation. Comprehensive simulation experiments prove the convergence of OMRA-DRL under various parameters. Moreover, it outperforms the five benchmark algorithms in terms of average cumulative reward and unfinished ratio of workloads, with an increase of over 15% in average cumulative reward and a decrease of about 5% in unfinished rate of workloads, demonstrating its effectiveness in achieving weighted cost minimization in edge collaborative networks.

关键词： Task offloading   Mobile edge computing   Trust   Blockchain   Digital twin   Game theory  

摘要： The rapid adoption of artificial intelligence (AI) across Internet of things (IoT) applications has intensified the demand for substantial computational resources. Due to the inherent resource limitations of IoT mobile devices, they are often unable to run most resource-intensive applications locally. Mobile edge computing (MEC) emerges as a promising solution, enabling task offloading to powerful edge servers. Although the technical aspects of task offloading are well-investigated in literature, the critical dimensions of security and trust concerns often remain under-explored. To address this gap, we introduce a blockchain-based trust management mechanism for digital Twin (DT) empowered MEC, facilitating informed offloading decision-making in task delegation and processing. The proposed trust management mechanism uses DT to enhance dynamic real-time task offloading. DT provides insights into the physical MEC environment and shares AI agent training data deployed on IoT devices, optimizing workload distribution. Our proposed trust management mechanism is based on an incentive cooperation model based on a hierarchical game model. Initially, we model cooperation among reliable servers as a hedonic coalition formation game, capturing edge servers' selfish behavior and aiming to maximize utilities. Subsequently, we model interactions between service-seeking devices and the coalition as a two-stage Stackelberg game, encouraging devices to delegate tasks to the most reliable coalition. The simulation results validate the effectiveness of our trustworthy approaches, demonstrating significant improvements in energy efficiency compared to existing related works. Our study shows up to a 60% increase in energy consumption fairness, while ensuring trust among nodes.

关键词： Mobile edge computing   Computational task scheduling   Offloading strategy   Multi-agent deep reinforcement learning  

摘要： The integration of mobile edge computing (MEC) into dynamic wireless ad hoc networks has intensified the challenges of computational resource scheduling, particularly under queue load constraints. Current research primarily focuses on global average metrics while neglecting fairness in resource allocation among devices across cross-time slot task scenarios. This oversight leads to significant disparities in the resources allocated to different devices, with some devices consistently lacking computational resources due to imbalanced scheduling. To address these limitations, we propose SEROS (Shared Exploration and Reward Optimization Strategy), a multi-agent reinforcement learning framework designed for cross-timeslot task scheduling in MEC environments. The method dynamically balances local optimization objectives with global collaboration through a weighted shared reward mechanism while enhancing training efficiency via hybrid sample trajectory utilization, enabling adaptive task offloading decisions. First, we construct a mobile edge computing model incorporating queue load constraints to address cross-timeslot task scheduling challenges, improving resource utilization for time-sensitive workloads through delayed optimization objectives. Second, we design a collaborative incentive mechanism based on global-local reward balancing and develop a sample trajectory-sharing scheme to accelerate policy convergence while preserving agent specialization. Simulation experiments validate the effectiveness of SEROS, demonstrating that compared with baseline methods, the proposed approach exhibits superior comprehensive performance in task completion rate improved by 7% and inter-device completion rate concentration enhanced by 40%, along with stability and task completion time.

关键词： User-centric network   Computation offloading   Resource allocation   Power control   Mobile edge computing   Reinforcement learning  

摘要： In mobile edge computing (MEC) over traditional cellular networks (TCN), users located at the cell edges are prone to severe edge interference and signal attenuation, leading to low throughput and even transmission interruptions. These edge effects significantly hinder the offloading of computational tasks from end devices to MEC servers, adversely affecting the user equipment (UE) experience. To address these issues and improve UE experience within the network, we design a user-centric network (UCN) structure comprising a three-tier network interconnected through communication links to ensure low latency and reliable transmission during computation offloading. For the computation offloading problem within UCN, we design a network model with dual-layer MEC servers that considers the competitive nature of UEs in real-world scenarios. This model is solved using a multi-agent reinforcement learning algorithm to optimize network strategies, achieving minimal long-term average total delay and power consumption. Experimental results demonstrate that the proposed scheme significantly reduces system power consumption, with a maximum reduction of 36.19% compared to other baseline algorithms. The scheme also achieves substantial reduction in the long-term average total delay by 39.5%. These algorithms indicate that the proposed approach offers considerable advantages in enhancing UE experience and reducing the energy consumption.

关键词： Machine learning   Latency optimization   Mobile edge computing   Fiber-wireless networks   Reinforcement Learning   Process  

摘要： The increasing quest for ultra-low latency in mobile edge computing (MEC) over fiber-wireless networks pose challenges in adapting to real-time demands under resource constraints. Congestion-prone methods of centralized learning and static routing waste resources. A machine learning-based latency optimization framework is proposed that integrates multiple advanced paradigms for proactive traffic management, intelligent routing, and efficient task offloading while ensuring data privacy. The framework encompasses self-supervised learning, federated learning, spatiotemporal graph neural networks (GNN), adaptive multi-agent reinforcement learning, and hypergraph transformers. These tools are useful for dynamic congestion-aware routing, accurate spatiotemporal traffic prediction, and optimal resource slicing for latencysensitive applications. Hypergraph transformers enable dynamic allocation of resources across network slices. The experiments shows significant performance improvements, 34-42 % reduction in end-to-end latency, 29-35 % faster task execution, 50 % less training time, better traffic prediction accuracy, and lower slice switching delays. This framework can underpin some of the most crucial low-latency applications like augmented reality and autonomous driving, providing a powerful solution for the next generation of MEC networks working under stringent perfor-mance and privacy constraints.. center dot Dynamic congestion-aware routing using adaptive multi-agent reinforcement learning center dot Proactive traffic prediction using spatiotemporal GNN center dot Federated, self-supervised learning to maximize resource slicing and delay-aware task offloading

关键词： Mobile edge computing   Simultaneously transmitting and reflecting   reconfigurable intelligent surface   Emergency communication   Computation offloading   Deep reinforcement learning  

摘要： Simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS) is a new wireless communication technology with bidirectional transmission. Unlike conventional reflective RIS, STAR-RIS offers unique advantages in multipath propagation, signal enhancement, and coverage. Through beamforming and intelligent signal processing, STAR-RIS provides users with additional transmission paths. In this context, this paper proposes a novel STAR-RIS-assisted unmanned aerial vehicle (UAV)-enabled mobile edge computing (U-MEC) emergency communication network. Considering the practical needs of post-disaster emergency communication, we formulate a problem of minimizing the maximum processing delay for all user tasks, while satisfying the battery energy constraints of UAV and guaranteeing the minimum transmission rate for users. To solve the problem, we comprehensively consider STAR-RIS beamforming, user task offload ratio, UAV flight trajectory, and resource allocation. Furthermore, we design an augmented Lagrangian-based reward-constrained twin delayed deep deterministic policy gradient (ALRCTD3) algorithm. The algorithm combines double-Q learning, state normalization, augmented Lagrangian relaxation method, and dual gradient descent strategy. Experimental results show that the proposed algorithm significantly outperforms existing deep reinforcement learning (DRL) algorithms and benchmark schemes in terms of performance.

关键词： Augmented reality   Energy efficiency   Hybrid optimization   Mobile edge computing   Mobility prediction   Task offloading  

摘要： Mobile edge computing enables the efficient execution of compute-intensive tasks by offloading them to edge servers. However, frequent user mobility in 5 G urban networks leads to increased latency, energy consumption, and resource wastage due to continuous handovers. To address these challenges, Energy Efficient Communication and Optimal Offloading Network, a framework is proposed that combines user mobility prediction and hybrid optimization for task offloading. Energy Efficient Communication and Optimal Offloading Network utilizes a modified Long Short-Term Memory model to predict user movement with high accuracy, achieving an accuracy improvement from 65 % to 95 % over ten iterations. Additionally, a Hybrid Grey Wolf Optimization Algorithm optimizes task allocation, resulting in a 30 % reduction in energy consumption and a 25 % improvement in server utilization compared to baseline methods. The framework achieves latency as low as 5 milliseconds for augmented reality tasks while maintaining scalability in high-traffic 5 G environments. The proposed model also outperforms baseline approaches in terms of task completion time, throughput, and communication efficiency, and it achieves a 94.5 % offloading success rate and 98 % augmented reality delay compliance. The proposed model provides a scalable and useful solution for real-time Augmented Reality by combining energy-constrained task allocation with mobility-aware predictions.

关键词： Increased Responsiveness   Internet of Things   Mobile Edge Computing   Task Computation Latency   Time Critical Responses  

摘要： INTRODUCTION: The Internet of Things (IoT) has transformed daily life by interconnecting digital devices via integrated sensors, software, and connectivity. Although IoT devices excel at real-time data collection and decision-making, their performance on complex tasks is hindered by limited power, resources, and time. To address this, IoT is often combined with cloud computing (CC) to meet time-sensitive demands. However, the distance between IoT devices and cloud servers can result in latency issues. OBJECTIVES: To mitigate latency challenges, Mobile Edge Computing (MEC) is integrated with IoT. MEC offers cloud-like services through servers located near network edges and IoT devices, enhancing device responsiveness by reducing transmission and processing latency. This study aims to develop a solution to optimize task offloading in IoT-MEC environments, addressing challenges like latency, uneven workloads, and network congestion. METHODS: This research introduces the Game Theory-Based Task Latency (GTBTL-IoT) algorithm, a two-way task offloading approach employing Game Matching Theory and Data Partitioning Theory. Initially, the algorithm matches IoT devices with the nearest MEC server using game-matching theory. Subsequently, it splits the entire task into two halves and allocates them to both local and MEC servers for parallel computation, optimizing resource usage and workload balance. RESULTS: GTBTL-IoT outperforms existing algorithms, such as the Delay-Aware Online Workload Allocation (DAOWA) Algorithm, Fuzzy Algorithm (FA), and Dynamic Task Scheduling (DTS), by an average of 143.75 ms with a 5.5 s system deadline. Additionally, it significantly reduces task transmission, computation latency, and overall job offloading time by 59%. Evaluated in an ENIGMA-based simulation environment, GTBTL-IoT demonstrates its ability to compute requests in real-time with optimal resource usage, ensuring efficient and balanced task execution in the IoT-MEC paradigm. CONCLUSION:

关键词： Super-resolution   Mobile edge computing   Video delivery   Quality of experience   Multi-user scenes  

摘要： As video traffic continues to dominate network data transmission, high playback latency has emerged as a critical bottleneck limiting the quality of streaming media services. While existing edge-assisted super-resolution methods mitigate latency in single-user scenes, they struggle to balance computational resource allocation and latency-sensitive task demands in multi-user scenes with constrained mobile edge computing (MEC) nodes. To this end, we propose Anableps, a multi-user MEC framework that integrates dynamic model selection and deep reinforcement learning-based scheduling to achieve efficient resource allocation and minimize latency. Specifically, we design a dynamic multi-granularity super-resolution loading model using optical flow field features to adaptively allocate light, medium, or heavy super-resolution models, optimizing both resource utilization and visual quality. Additionally, we design a dynamic multi-task scheduling algorithm based on deep Q-Network that dynamically adjusts system load and task urgency through adaptive penalty factors, ensuring efficient resource prioritization. Experimental results demonstrate that Anableps improves QoE by 12.7% compared to state-of-the-art methods, while enhancing video smoothness and reducing re-buffering time by 23.4% and 18.7%, respectively, which further highlights its effectiveness in optimizing streaming performance in resource-constrained environments.

关键词： Mobile edge computing   Zero-shot task scheduling   Self-supervised learning   Deep Reinforcement Learning   Soft Actor-Critic  

摘要： The rising need for swift response times makes it essential to use computing resources and network capacities efficiently at the edges of the networks. Mobile Edge Computing (MEC) handles this by processing user data near where it is generated rather than always relying on remote cloud centres. Yet, scheduling tasks under these conditions can be difficult because workloads shift, resources vary, and network performance is unstable. Traditional scheduling strategies often underperform in such rapidly changing settings, and even Deep Reinforcement Learning (DRL) solutions usually require extensive retraining whenever they encounter unfamiliar tasks. This paper proposes a self-supervised DRL framework for zero-shot task scheduling in MEC environments. The system integrates self-supervised learning to generate task embeddings, enabling the model to classify tasks into clusters based on resource requirements and execution complexity. A Soft Actor-Critic (SAC)-based scheduler then optimally assigns tasks to MEC nodes while dynamically adapting to network conditions. The training process combines contrastive learning for task representation and policy optimization to enhance scheduling decisions. Simulations demonstrate that the proposed approach reduces task completion time by up to 22 %, lowers energy consumption by 29 %, and improves latency by 18 % over baseline methods.