Huilin Deng, Hongchen Luo, Yue Zhu, Long Li, Zhuoyue Chen, Xinghao Zhao, Ming LI, Chuyang Zhao, Jihai Zhang, MengChang Wang, Yang Cao, Yu Kang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

With the widespread application of the Internet of Things (IoT), computing tasks on the terminal side have surged. Traditional cloud computing models, constrained by high network latency and overloaded central servers, can no longer effectively meet the dual requirements of real-time responsiveness and energy efficiency. The cloud–edge–device collaborative architecture, by enabling distributed resource scheduling, offers a promising solution to reduce both latency and energy consumption. However, optimizing carbon emissions under dynamic operating conditions remains a pressing and unresolved challenge. This paper proposes a carbon-aware dynamic scheduling framework for cloud–edge–device systems, which accounts for the stochastic nature of task arrivals, heterogeneous computing capabilities, and varying carbon intensity across devices and locations. A multi-layer carbon emission model is developed, and the long-term carbon minimization objective is formulated as a stochastic optimization problem. Using the Lyapunov drift-plus-penalty method, the problem is transformed into a tractable deterministic optimization framework, upon which a Carbon-Efficient Computation Offloading (CECO) algorithm is designed. CECO jointly optimizes local computation frequency, data transmission rate, and edge resource allocation to dynamically balance task queue stability and carbon emission intensity. Theoretical analysis and simulation results validate that the proposed algorithm significantly reduces system-level carbon emissions while maintaining quality of service, demonstrating strong potential for enabling green computing in intelligent distributed environments.

Multi-access Edge Computing (MEC) can effectively process Internet of Things (IoT) data by transferring computing intensive tasks to edge servers, and has become an effective mechanism to meet the growing demand for computing. The flexible Unmanned Aerial Vehicle (UAV) and High-Altitude Platform (HAP) with powerful resources working together can significantly improve the efficiency of edge computing system. This paper investigates the resource allocation and trajectory optimization problems in HAP-UAV-MEC system with a Non-Orthogonal Multiple Access (NOMA) communication scenario. By utilizing Wireless Power Transfer (WPT) technology to provide energy support for UAV, we jointly optimize UAV trajectories, resource allocation, and offloading decisions to minimize the energy cost of IoT devices and the energy cost of UAV. This problem is described as a multi-stage Mixed Integer Nonlinear Program ming (MINLP) problem. A Deep Deterministic Policy Gradient (DDPG)-Attention-based Resource Allocation and Trajectory Optimization (DART) algorithm combining Deep Reinforcement Learning (DRL) and Lyapunov optimization techniques is proposed to address this issue. DART algorithm utilizes the Lyapunov technique to transform the multi-stage MINLP problem into a deterministic optimization problem, and decomposes the original problem into four parallel subproblems. Through DDPG-attention algorithm based on reinforcement learning and deep learning attention mechanisms, we solve the problems of trajectory optimization and offloading decision. Meanwhile, for remaining subproblems related to resource allocation, convex optimization is used to solve them. The experimental results verify that the DART algorithm can significantly reduce the total cost while ensuring system stability and performance.

Due to the limited battery capacity and computational resources of mobile devices, computation-intensive tasks generated by mobile devices can be offloaded to edge servers for processing. This paper investigates the multi-user task offloading and resource pricing issues in Autonomous aerial vehicle (AAV)-assisted Multi-Access Edge Computing (MEC) systems. The optimization objectives is optimizing the utility of the server and the utility of the Edge Users (EUs), with decision variables encompassing the offloading strategies of EUs and the pricing strategies of the server. We divide the entire optimization problem into two parts. When optimizing the server's utility, server energy consumption is a crucial metric; hence, in the first part, we formulate the user allocation problem with the goal of minimizing the server's overall energy consumption. Utilizing game theory, we transform the user allocation problem into a multi-user non-cooperative game and prove the existence of a Nash Equilibrium (NE). The Game-based User Allocation (GBUA) algorithm is proposed to obtain the user allocation strategy. After addressing the user allocation problem, we consider the simultaneous optimization of both server and EUs utility. Therefore, in the second part, we model the server and EUs's engagement using the Stackelberg game model and employ backward induction to verify the presence of a Stackelberg Equilibrium (SE). Additionally, we propose the Resource Pricing and Task Offloading (RPATO) algorithm, based on game theory, to obtain the SE solution. Finally, extensive experiments are conducted to validate the effectiveness of the proposed algorithms, and numerous comparative algorithms are tested to prove the advancement and innovation of our proposed algorithms.

In UAV-assisted Multi-access Edge Computing (MEC) systems, UAV trajectories and resource pricing directly determine system utility - improper UAV positioning will lead to increased delay and energy costs for users, while inappropriate pricing will result in insufficient task offloading or UAV overload. This paper aims to optimize UAVs' trajectories, pricing strategies and users' offloading decisions, enabling UAVs to move to suitable locations and provide computing offloading services at appropriate prices, thereby reducing delay and energy, and enhancing the utilities of both UAV and users. The user utility specifically consist of throughput, energy consumption, delay penalties, and expenditures on purchasing computing resources. The UAV's utility consists of computing energy consumption, delay penalties, and revenue from selling computing resources as incentives. We formulate the pre-offloading problem to solve users' offloading selection, and transform the problem into a multi-user non-cooperative game using game theory while proving the existence of Nash equilibrium. Then we model the interaction between UAV and users using Stackelberg game model and prove the existence of Stackelberg equilibrium. We use multi-agent deep reinforcement learning (MADRL) and propose DRL and Game Theory-based Trajectory Optimization and Task Offloading (DGTT) algorithm to solve UAVs' trajectories and obtain the Stackelberg equilibrium solution. We sequentially solve for the pricing strategy and offloading decisions, thereby obtaining the optimized utilities of both UAV and users. Finally, we conduct simulation experiments to verify the feasibility of DGTT algorithm, along with comparative experiments that demonstrate our proposed DGTT algorithm's excellent performance in optimizing UAV and user utilities, while reducing system energy consumption and delay.

We study the stochastic combinatorial semi-bandit problem with unrestricted feedback delays under merit-based fairness constraints. This is motivated by applications such as crowdsourcing, and online advertising, where immediate feedback is not immediately available and fairness among different choices (or arms) is crucial. We consider two types of unrestricted feedback delays: reward-independent delays where the feedback delays are independent of the rewards, and reward-dependent delays where the feedback delays are correlated with the rewards. Furthermore, we introduce merit-based fairness constraints to ensure a fair selection of the arms. We define the reward regret and the fairness regret and present new bandit algorithms to select arms under unrestricted feedback delays based on their merits. We prove that our algorithms all achieve sublinear expected reward regret and expected fairness regret, with a dependence on the quantiles of the delay distribution. We also conduct extensive experiments using synthetic and real-world data and show that our algorithms can fairly select arms with different feedback delays.