The rapid advancement of Artificial Intelligence (AI) has introduced Deep Neural Network (DNN)-based tasks to the ecosystem of vehicular networks. These tasks are often computation-intensive, requiring substantial computation resources, which are beyond the capability of a single vehicle. To address this challenge, Vehicular Edge Computing (VEC) has emerged as a solution, offering computing services for DNN-based tasks through resource pooling via Vehicle-to-Vehicle/Infrastructure (V2V/V2I) communications. In this paper, we formulate the problem of joint DNN partitioning, task offloading, and resource allocation in VEC as a dynamic long-term optimization. Our objective is to minimize the DNN-based task completion time while guaranteeing the system stability over time. To this end, we first leverage a Lyapunov optimization technique to decouple the original long-term optimization with stability constraints into a per-slot deterministic problem. Afterwards, we propose a Multi-Agent Diffusion-based Deep Reinforcement Learning (MAD2RL) algorithm, incorporating the innovative use of diffusion models to determine the optimal DNN partitioning and task offloading decisions. Furthermore, we integrate convex optimization techniques into MAD2RL as a subroutine to allocate computation resources, enhancing the learning efficiency. Through simulations under real-world movement traces of vehicles, we demonstrate the superior performance of our proposed algorithm compared to existing benchmark solutions.

Fog computing is an energy-efficient and cost-effective paradigm to help alleviate the pressure of resource-constrained mobile devices (MDs) running computation-intensive applications. In this article, we investigate the joint task partitioning and power control problem in a fog computing network with multiple MDs and fog devices (FDs), where each MD has to complete a periodic computation task under the constraints of delay and energy consumption. Each task can be partitioned into multiple subtasks and offloaded to the FDs according to the task partition strategy and transmission power strategy to reduce task execution delay and energy consumption. To this end, we present a multiagent deep deterministic policy gradient (MADDPG)-based task offloading algorithm for MDs to maximize the long-term system utility including the execution delay and energy consumption. Each MD inputs the local information, e.g., the task requirements, the available communication, and computation resources of the FDs, the computation resources, and the battery level of the MD into a distributed actor network to generate a task offloading policy, while a centralized critic network is used to update the weights of the actor networks to improve offloading performance. Numerical simulation results demonstrate the effectiveness of the proposed scheme in improving the system utility, reducing the average execution delay as well as the average energy consumption.

Federated learning (FL) has emerged as a promising paradigm for enabling collaborative training of machine learning models while preserving data privacy. However, the massive heterogeneity of data and devices, communication constraints, and security threats pose significant challenges to its practical implementation. This paper provides a system review of the state-of-the-art techniques and future research directions in FL, with a focus on addressing these challenges in resource-constrained environments by a cloud–edge–end collaboration FL architecture. We first introduce the foundations of cloud–edge–end collaboration and FL. We then discuss the key technical challenges. Next, we delve into the pillars of trustworthy AI in the federated context, covering robustness, fairness, and explainability. We propose a dimension reconstruction of trusted AI and analyze the foundations of each trustworthiness pillar. Furthermore, we present a lightweight FL framework for resource-constrained edge–end devices, analyzing the core contradictions and proposing optimization paradigms. Finally, we highlight advanced topics and future research directions to provide valuable insights into the field.

Internet of Underwater Things (IoUT) consists of numerous sensor nodes distributed in an underwater area for sensing, collecting, processing information, and sending related messages to the data processing center. However, the characteristics of the underwater environment will bring strict limitations on communication coverage and power scarcity to IoUT networks. Applying cooperative communications to IoUT networks can expand the communication range and alleviate power shortages. In this article, we investigate the cooperative communication problem in a power-limited cooperative IoUT system and propose a reinforcement learning-based underwater relay selection strategy. Specifically, we first determine the optimal transmit powers of the source node and the selected underwater relay to maximize the end-to-end signal-to-noise ratio of the system. Then, we formulate the underwater cooperative relaying process as a Markov process and apply reinforcement learning to obtain an effective underwater relay selection strategy. The simulation results show that the performance of the proposed scheme outperforms that of the equal transmit power settings under the same conditions. In addition, the proposed deep Q-network-based underwater relay selection strategy improves the communication efficiency compared with the Q-learning-based strategy, and the number of iterations needed for convergence can be effectively reduced.

The growth of smart vehicles and computation-intensive applications poses new challenges in providing reliable and efficient vehicular services. Offloading such applications from vehicles to mobile edge cloud servers has been considered as a remedy, although resource limitations and coverage constraints of the cloud service may still result in unsatisfactory performance. Recent studies have shown that exploiting the unused resources of nearby vehicles for application execution can augment the computational capabilities of application owners while alleviating heavy on-board workloads. However, encouraging vehicles to share resources or execute applications for others remains a sensitive issue due to user selfishness. To address this issue, we establish a novel computation offloading marketplace in vehicular networks where a VCG-based reverse auction mechanism utilizing integer linear programming (ILP) problem is formulated while satisfying the desirable economical properties of truthfulness and individual rationality. As ILP has high computation complexity which brings difficulties in implementation under larger and fast changing network topologies, we further develop an efficient unilateral-matching-based mechanism, which offers satisfactory sub-optimal solutions with polynomial computational complexity, truthfulness and individual rationality properties as well as matching stability. Simulation results show that, as compared with baseline methods, the proposed unilateral-matching-based mechanism can greatly improve the system efficiency of vehicular networks in all traffic scenarios.

Vehicular Clouds (VCs) are modern platforms for processing of computation-intensive tasks over vehicles. Such tasks are often represented as Directed Acyclic Graphs (DAGs) consisting of interdependent vertices/subtasks and directed edges. However, efficient scheduling of DAG tasks over VCs presents significant challenges, mainly due to the dynamic service provisioning of vehicles within VCs and non-Euclidean representation of DAG tasks’ topologies. In this paper, we propose a Graph neural network-Augmented Deep Reinforcement Learning scheme (GA-DRL) for the timely scheduling of DAG tasks over dynamic VCs. In doing so, we first model the VC-assisted DAG task scheduling as a Markov decision process. We then adopt a multi-head Graph ATtention network (GAT) to extract the features of DAG subtasks. Our developed GAT enables a two-way aggregation of the topological information in a DAG task by simultaneously considering predecessors and successors of each subtask. We further introduce non-uniform DAG neighborhood sampling through codifying the scheduling priority of different subtasks, which makes our developed GAT generalizable to completely unseen DAG task topologies. Finally, we augment GAT into a double deep Q-network learning module to conduct subtask-to-vehicle assignment according to the extracted features of subtasks, while considering the dynamics and heterogeneity of the vehicles in VCs. Through simulating various DAG tasks under real-world movement traces of vehicles, we demonstrate that GA-DRL outperforms existing benchmarks in terms of DAG task completion time.