With the rapid development of Generative Artificial Intelligence (GAI), large-scale AI-Generated Content (AIGC) has been widely produced, raising critical challenges in trustworthy copyright management. Blockchain-based copyright registration or trading have become a research hotspot, but existing solutions focus on isolated stages and fail to support the full lifecycle of AIGC content, while copyright management performance, infringement detection capability, and copyright query efficiency remain challenging. To address these challenges, we designed a dual-layer blockchain framework for full-lifecycle AIGC copy-right management, which supports coordinated copyright registration, verification, trading, and traceability. The proposed framework adopts a dual-layer architecture with a main chain and multiple sub-chains, and integrates sharding with a Directed Acyclic Graph (DAG) parallel ledger to improve system scalability. Specifically, a Perceptual Hash (pHash)-based similarity detection method is introduced for copyright registration to identify plagiarism and unauthorized duplication; a hybrid indexed sharded query mechanism is designed for efficient and verifiable copyright verification; and cryptographic techniques together with zero-knowledge proofs are incorporated to enable secure and non-repudiable copyright trading. Experimental results show that the designed framework delivers about 1.1× higher throughput and achieves roughly a 29× reduction in transaction latency compared with single-chain blockchains, while the proposed query mechanism reduces query latency by up to 56× across different shard scales. These results validate the capability of the proposed framework to support secure, efficient, and scalable AIGC copyright management.

Open Radio Access Network (O-RAN) supports heterogeneous service coexistence through functional splitting and open interfaces, enabling traffic steering via functional orchestration and resource scheduling. However, existing studies focus on known traffic patterns and lack the ability to anticipate dynamic service demands in advance. Isolated optimization of orchestration and scheduling fails to ensure End-to-End (E2E) latency. The varying time scales and vast solution space further complicate the joint optimization. To address this, we propose a traffic twin-enabled orchestration and scheduling multi-timescale joint optimization scheme. Explicitly, we design a spatiotemporal attention-assisted Time Series Generative Adversarial Network (TimeGAN) traffic twin model (STAG-TD) to capture unknown traffic patterns. Based on twin results, we formulate a joint optimization problem and design a dual-timescale algorithm framework, including propose a Task Decomposed Dueling Double Deep Q-Network (TD3QN) algorithm to handle large-timescale orchestration, and use a Penalty-based Particle Swarm Optimization (PPSO) algorithm to manage small-timescale scheduling. Our scheme achieves a predictive joint optimization to reduce the transmission latency of services. Extensive results show our scheme outperforms state-of-the-art methods, reducing E2E latency by over 39% and increasing throughput by over 14.9%. The highly consistent results between real and twin data also demonstrate the effectiveness of the traffic twin model.

Artificial Intelligence Generated Content (AIGC) plays a key role in shaping the emerging metaverse ecosystem through its ability to efficiently and automatically generate large scale, personalized content. While high-quality AIGC generation under a cloud-edge-end three-layer architecture has attracted significant research attention, existing approaches often overlook trust challenges throughout the AIGC service lifecycle namely, in model provision, Service Provider (SP) selection, and product transaction. To address these issues, we introduce blockchain technology and propose a cloud-edge collaborative, blockchain oriented AIGC service architecture (CEAIGC). This architecture ensures secure and trustworthy interactions among AIGC model providers, SPs, and users. Specifically, we design embedded watermark coding rules for AIGC models and use blockchain to verify consistency between cloud and edge models, providing a reliable foundation for SPs. To further support trustwor thy SP selection, we formulate a multi-objective optimization problem that considers user utility, SP reputation, and energy consumption. We then propose a diffusion-model-enhanced Deep Reinforcement Learning (DRL) algorithm (DMA3C) to optimize SP selection and adaptively match metaverse user needs, enabling reliable, low-latency AIGC inference at the edge. To overcome blockchain performance bottlenecks, we employ a smart contract engine to establish state channels between transaction users. This enables efficient, secure, and atomic off-chain transfers of AIGC product ownership and service fees. Extensive experiments demonstrate that CEAIGC improves system throughput by 2.38×, and the proposed DMA3C algorithm achieves performance gains of 11.4% to 28.6% compared to other DRL-based approaches.

Celluar traffic forecasting is critical for optimizing resources, balancing loads, and reducing costs in cellular networks. However, newly established base stations often suffer from data scarcity due to the lack of historical traffic data, and the varying traffic demands across regions and times further complicate accurate predictions. While deep learning models typically perform well in data-rich environments, their accuracy drops significantly in data-scarce areas. To tackle this challenge, we propose a Deep Attention-based Graph Transfer Network(DAGTN), a novel framework that leverages cross-domain transfer learning. The framework includes a spectral clustering algorithm based on Dynamic Time Warping (DTW) to group similar base stations, enabling effective knowledge transfer and capturing spatial dependencies. We also introduce DSAN, a deep temporal prediction model that integrates attention mechanisms and sampling strategies to capture traffic patterns more accurately. Additionally, Generative Adversarial Networks (GANs) are used for domain adaptation, aligning the distribution between source and target domains while maintaining data privacy. Our experiments show that DAGTN outperforms existing methods, improving prediction accuracy by 8.7% in RMSE, MAE, and R2. Ablation studies further validate the contribution of each component.

With the development of immersive applications, eXtended Reality (XR) has emerged as a key application in scenarios such as Industrial Internet of Things (IIoT) and smart healthcare, where the demand for low-latency and high-bandwidth network performance is particularly urgent. To meet the low-latency requirements of multiple users in multiple XR services, service caching and wireless resource scheduling have become essential approaches. However, most existing studies focus on optimizing only one of these two aspects, ignoring the coupling relationship between the two, which makes it difficult to comprehensively improve the quality of XR services. To address this challenge, we propose a joint service caching and wireless resource scheduling method for low-latency XR delivery. By comprehensively considering user requests and service features, we formulate a joint optimization model with the objective of minimizing end-to-end latency. To solve this model, we design a Transformer-Enabled Multi-Agent ActorCritic (TMAC) algorithm. Specifically, we first introduce Graph Neural Network (GNN) to capture network features. Then, we model each XR service as an agent and design a Transformer-based actor network to make the service caching decision, while incorporating a global actor network based on Hypergraph Neural Network (HGNN) to generate resource scheduling decisions. This method achieves collaborative optimization of service caching and wireless resource scheduling while enhancing system flexibility and decision-making efficiency. Compared to various baseline methods, the proposed algorithm reduces the average service latency by approximately$\mathbf{1 5. 8 8} \boldsymbol{\%} \mathbf{- 3 1. 2 9} \boldsymbol{\%}, \mathbf{2 6. 7 3} \boldsymbol{\%} \mathbf{4 5. 0 9 \%,} \mathbf{2 9. 9 5 \%} \boldsymbol{-} \mathbf{4 9. 4 8 \%}$respectively, significantly improving the QoS (Quality of Service) and low-latency guarantee capability of XR service delivery in multi-user, multi-service scenarios.

To enable intelligent and low-cost End-to-End (E2E) network service deployment and Service Level Agreement (SLA) quality management in the two-level Cloud Radio Access Network (C-RAN), this paper studies a DTN-based SLA quality closed-loop management scheme, which mainly includes acquisition module, base module, deployment module, and monitoring module. The deployment module is responsible for constructing the service deployment optimization model with the goal of minimizing the average E2E delay of packets, and quickly obtain deployment decisions through a Weighted GraphSAGE (WGraphSAGE)-assisted Double Deep Q-network (DDQN)-based two-stage service deployment (WDTSD) algorithm. The monitoring module uses the state monitoring model based on Bayesian Convolutional Neural Network (BCNN) to complete the abnormal detection of physical devices. The modular closed-loop interaction provides a virtual environment for network service deployment, verification, monitoring, and policy revision, achieving SLA quality assurance. Extensive results validate the effectiveness of the WDTSD algorithm, state monitoring model, and DTN. WDTSD outperforms existing solutions in terms of memory overhead, computing speed, E2E delay, and service access ratio. The state monitoring model has better performance in indicators such as accuracy. The results under different data acquisition periods show that the service deployment effect is better when the DTN is closer to the physical network.

The distribution of computing resources in the Computing Power Network (CPN) is uneven, leading to an imbalance in resource supply and demand within domains, necessitating cross-domain resource scheduling. To address the cross-domain resource scheduling challenge in CPN, this paper presents an Improved Multi-Agent Actor-Critic (IMAAC) resource scheduling approach leveraging Data Processing Unit (DPU) offloading. Initially, we introduce a cross-domain resource scheduling architecture tailored for CPN by leveraging DPU offloading. Specifically, we delegate certain functionalities of the Multi-Agent Deep Reinforcement Learning (MADRL) Agent to DPUs, aiming to mitigate communication costs incurred during the generation of cross-domain scheduling decisions. Second, we introduce the parallel experience ensemble and multi-head attention mechanism in the Multi-Agent Actor-Critic (MAAC) framework to compress the state-space dimensionality of agent association across domains. Finally, we introduce the parallelized dual-policy network structure to mitigate training instability and convergence challenges within the actor and critic networks. Experimental results showcase that IMAAC achieves noteworthy reductions of 5.98%~13.56%, 23.54%~33.55%, and 41.17%~58.88% in total system delay, energy consumption, and the number of discarded tasks, respectively, compared to benchmark experiments.

As a new trusted data sharing pattern with privacy protection, the integration mechanism of blockchain and Federated Learning has attracted extensive attention. Generally, this mechanism uses blockchain technology to supervise the original data and calculation results, which ignores the supervision of the Federated Learning model and computing process. Therefore, we introduce the concepts of the sandbox and state channel to construct a new data privacy sharing paradigm via Blockchain and Federated Learning. Under this paradigm, we use state channel to connect Blockchain and Federated Learning. And state channel is used to create a “trusted sandbox” to instantiate Federated Learning tasks in the trustless edge computing environment. Meanwhile, we also mainly solve problems about data privacy sharing in Federated Learning and system performance degradation caused by data quality. The simulation results show that the proposed method has better performance and efficiency than the traditional data sharing method.

The development of edge computing has alleviated the problem of limited vehicular computing capabilities in VANET. The vehicular edge computing (VEC) provide resources for the implementation of multiple intelligent services. However, the mobility of vehicles and the diversity of edge computing nodes pose huge challenges for service offloading. Deep reinforcement learning (DRL) in artificial intelligence (AI) is an effective technology to solve such challenges. Based on this scenario, we first introduce a software-defined vehicular networks (SDV) architecture that takes full advantage of the characteristics of SDN technology and can effectively and dynamically obtain a global view in VANET to facilitate the management of resources in the network. Then, we propose a new intelligent service offloading decision model, which introduces the Deep Deterministic Policy Gradient (DDPG) algorithm in DRL to solve the joint optimization of service offloading with multiple constraints. Simulation results show that the DDPG-based service offloading model has better performance and better stability than similar algorithms.