Web platforms are reshaping resource allocation in distributed energy networks globally, from off-grid communities to lunar bases. Algorithmic decision-makers face the fundamental challenge of fairly distributing scarce resources among heterogeneous stakeholders. Traditional approaches assume complete rationality with perfect information and unlimited computation, yet distributed networks only permit local observation, requiring fairness to emerge from individual strategic interactions. Centralized optimization fails due to exponential complexity, rule-based methods cannot adapt to disruptions, and existing platforms translate economic inequality into energy access inequality. Recognizing the unattainability of complete rationality necessitates bounded rationality: pursuing provably convergent satisficing solutions under incomplete information and limited computation, translating natural language ethics into computable constraints, and designing incentives so self-interested behavior satisfies fairness at equilibrium. We propose a unified semantic-game-distributed framework. Large language models map ambiguous ethical principles into game-theoretic parameters through semantic parameterization, with contrastive learning ensuring semantic consistency and temporal stability. A two-layer Stackelberg game implements incentive design: the platform signals through differentiated pricing while nodes optimize locally, enabling fairness to emerge from equilibrium. Distributed asynchronous iteration achieves global convergence through local communication, with cognitive models adaptively adjusting step sizes and differential perturbation preserving privacy. Theoretical analysis establishes equilibrium existence and convergence guarantees, while extreme scenarios validate robustness under information scarcity and high uncertainty.

ABSTRACT Graph neural networks enhance collaborative signals through multi‐hop message passing, achieving promising results in recommender systems. Nevertheless, several challenges still remain. Most existing methods rely solely on user‐item interactions, which are insufficient to capture users' fine‐grained preferences. Although some approaches exploit aspect information to alleviate this issue, they typically treat aspects as auxiliary information and fail to explicitly integrate them into user (item) embeddings. Additionally, these methods generally treat all unobserved items as negative samples and employ a uniform sampling strategy. However, due to the inherent popularity bias of items, such a negative sampling strategy is suboptimal for effective model training. To address the above issues, we propose a novel method, AEN‐GCL, which leverages aspect information to help construct the initial embeddings of users (items) and guide negative sampling, while incorporating a self‐supervised auxiliary learning task. Specifically, we first employ a pre‐trained Sentence‐BERT to compute the similarities between reviews and predefined seed sentences of each aspect. The similarities derived from the reviews of each user (item) are aggregated to form an aspect representation, which is combined with category information to construct the initial embedding. Next, we compute the aspect‐level similarity between the user, the positive item and candidate negatives generated by an adaptive category‐based strategy. The user‐aware and item‐aware similarities are jointly used to guide negative sampling. In addition, to alleviate sparse supervision signals, we incorporate edge‐dropout‐based graph contrastive learning to enhance model training. Comprehensive experiments on three public datasets demonstrate that the proposed AEN‐GCL achieves competitive performance compared with state‐of‐the‐art methods.

The Internet of Vehicles (IoV), a critical large-scale Internet of Things (IoT) application, faces a fundamental sustainability challenge stemming from the lifecycle mismatch between long-duration vehicular hardware and rapidly evolving software. This mismatch creates a widening “computational divide,” where aging vehicles with limited onboard resources cannot support modern data-intensive applications, thereby fragmenting the ecosystem and undermining its collective intelligence. To address this systemic issue, this paper proposes a novel lifecycle-aware task offloading framework. Instead of treating vehicular heterogeneity as a static liability, our framework transforms it into a dynamic, cooperative resource-sharing opportunity. At the core of this framework is a Deep Reinforcement Learning (DRL) agent deployed on resource-constrained vehicles, which orchestrates task offloading decisions to co-optimize for latency and energy consumption. A key innovation is the “Performance Capacity” metric, a multi-dimensional and predictive state assessment mechanism. This metric enables informed decision-making by intelligently fusing a vehicle’s static hardware profile, its dynamically predicted future workload, and its cooperation reputation. Furthermore, an integrated credit-based incentive mechanism is designed to ensure the long-term economic viability of this cooperative ecosystem. Simulation results demonstrate that our framework significantly reduces task latency and energy consumption, particularly under high-load conditions, offering a scalable and sustainable solution for the continuous evolution of heterogeneous IoV systems.

Traditional Internet of Things (IoT) technologies in complex pipeline networking systems of industrial automation face limitations in sensor data acquisition, making it challenging to comprehensively capture the complex distributions of multi-physical fields, such as pressure, temperature, and velocity. This restricts holistic system analysis and optimal decision-making capabilities. However, accurate and efficient prediction of these multi-physical fields is essential for intelligent decision-making and optimization in IoT-enabled industrial systems. Traditional Computational Fluid Dynamics (CFD) methods deliver high fidelity but are computationally expensive and unsuitable for real-time monitoring and control scenarios characteristic of IoT environments. Recent AI-based generative methods, including Transformers and generative adversarial networks (GANs), improve computational efficiency but often suffer from overly smooth predictions and training instability, limiting their effectiveness for precise industrial IoT applications. Motivated by the outstanding performance of diffusion models in generative tasks, we propose CFDiff, a diffusion-based architecture designed explicitly for predicting complex multi-physical fields in IoT-based industrial pipeline applications. By introducing the simulation-free flow-map, a single-channel orientation field that encodes the inlet-to-outlet direction, CFDiff leverages a generative diffusion process combined with a lightweight Cross-Attention Fusion (CAF) module, effectively integrating sparse and multimodal data to generate high-fidelity field distributions. Experimental results demonstrate that CFDiff significantly outperforms existing state-of-the-art methods, reducing prediction errors by approximately 41.6% across various scenarios relevant to industrial IoT. Comprehensive ablation studies further confirm the efficacy of each proposed component, positioning CFDiff as a robust and practical solution to enhance the accuracy, safety, and efficiency of IoT-based monitoring and predictive maintenance systems.

Anomaly detection in Industrial Internet of Things (IIoT) systems is crucial for ensuring operational safety and product quality. However, challenges such as heterogeneous sensor data, dynamic network structures, and complex spatio-temporal dependencies hinder existing methods. To address these issues, we propose a novel Transformer-based anomaly detection framework designed for dynamic IIoT networks. Our framework integrates self-supervised graph contrastive learning with multi-center Support Vector Data Description (SVDD) for the first time. Specifically, we construct temporal graph snapshots from normal time windows. Graph Attention Networks (GAT) are used to extract structural features, while a lightweight Transformer with temporal window attention captures long-range dependencies among sensor sequences, producing robust spatio-temporal node representations. To improve the model’s generalization without requiring labeled anomalies, a self-supervised contrastive loss is introduced. The resulting node embeddings are clustered via k-Medoids, and an SVDD sub-model is trained per cluster to define multiple hyperspheres representing normal behavior. During inference, the proximity of test representations to these hyper-spheres enables accurate and interpretable anomaly detection. Extensive experiments on real-world IIoT datasets, including SWaT and WADI, show our method outperforms state-of-the-art baselines in AUC and F1 score, demonstrating its effectiveness and practical value for IIoT security applications.

Deep reinforcement learning (DRL) has shown significant potential in adaptive traffic signal control (ATSC) by adapting to real-time traffic conditions. However, controlling multiple intersections faces challenges, mainly due to the isolated actions of agents and non-stationary caused by other intersections. To address these issues, this paper proposes a novel knowledge collaboration-based actor-critic policy gradient (KCACPG) method to achieve cooperative traffic scheduling across multiple intersections. KCACPG includes a knowledge collaboration learning mechanism that allows heterogeneous agents to exchange knowledge across experience tuples, achieving globally optimal decision-making and coordination. KCACPG also integrates an off-policy prioritized experience replay mechanism to improve knowledge reuse efficiency and reduce the negative impact of knowledge transfer. Simulation results show that KCACPG converges quickly, generalizes to fluctuant traffic and load well, improves the network throughput by up to 17.8%, and reduces the pressure imbalance by up to 11.6% compared with the existing collaborative methods. The proposed method has significant implications for intelligent transportation systems and smart cities.

Abstract Diffusion models can generate high‐quality images and have attracted increasing attention. However, diffusion models adopt a progressive optimization process and often have long training and inference time, which limits their application in realistic scenarios. Recently, some latent space diffusion models have partially accelerated training speed by using parameters in the feature space, but additional network structures still require a large amount of unnecessary computation. Therefore, we propose the Contour Wavelet Diffusion method to accelerate the training and inference speed. First, we introduce the contour wavelet transform to extract anisotropic low‐frequency and high‐frequency components from the input image, and achieve acceleration by processing these down‐sampling components. Meanwhile, due to the good reconstructive properties of wavelet transforms, the quality of generated images can be maintained. Second, we propose a Batch‐normalized stochastic attention module that enables the model to effectively focus on important high‐frequency information, further improving the quality of image generation. Finally, we propose a balanced loss function to further improve the convergence speed of the model. Experimental results on several public datasets show that our method can significantly accelerate the training and inference speed of the diffusion model while ensuring the quality of generated images.

Facial expressions are important for conveying information in human interactions. The diffusion model can generate high-quality images for clearer and more discriminative faces, but its training and inference time is often prolonged, hampering practical application. Latent space diffusion models have shown promise in speeding up training by leveraging feature space parameters, but they require additional network structures. To address these limitations, we propose a contour wavelet diffusion model that accelerates both training and inference speeds. We use a contour wavelet transform to extract components from images and features, achieving substantial acceleration while preserving reconstruction quality. A normalised random channel attention module enhances the quality of generated images by focusing on high-frequency information. We also include a reconstruction loss function to enhance convergence speed. Experimental results demonstrate the effectiveness of our approach in boosting the training and inference speeds of diffusion models without sacrificing image quality. Fast generation of facial expressions can provide a smoother and more natural user experience, which is important for real-time applications. In addition, the increase in inference speed can save the use of computational resources, reduce system cost and improve energy efficiency, which is conducive to promoting the development and application of this technology.