Song Wang

The rapid proliferation of Internet of Things (IoT)-enabled cameras has driven increasing demand for low-latency, intelligent video understanding in real-world applications such as smart cities and industrial automation. While multimodal large language models (MLLMs) offer unprecedented capabilities in semantic reasoning and natural language-based video comprehension, their deployment in latency-sensitive IoT environments remains challenging due to high computational costs and sequential decoding bottlenecks. Moreover, conventional edge–cloud video analysis frameworks often rely on semantic-agnostic frame sampling, leading to information loss or redundant data transmission. In this article, we propose $SEC$ , a semantic-aware edge–cloud collaborative framework for efficient and accurate video analysis. $SEC$ introduces a task-aware key frame selection mechanism at the edge to maximize semantic relevance while minimizing bandwidth usage and a novel adaptive speculative decoding (SPD) framework with tree-based parallel generation on the cloud to accelerate MLLM inference. Extensive experiments under realistic edge–cloud deployment settings on four video understanding benchmarks demonstrate that the proposed $SEC$ framework achieves superior performance, significantly reducing the end-to-end inference latency while improving the accuracy, a rare win-win in latency-critical IoT systems.

We investigate a hydrogen sensor based on a one-dimensional core-shell nanoparticle chain. The nanoparticles are composed of silver cores coated with palladium (Pd) shells. The distance between adjacent particles is larger than three times the unit radius. This configuration allows for the analysis of the optical response of the designed structure using both coupled dipole theory (CD) and the finite-difference time-domain (FDTD) method. Leveraging the change in the dielectric constant of Pd before and after hydrogen absorption, combined with the collective effects of surface plasmons, the presence of H ₂ is detected through differences in the absorption cross-sections. Results show that by constructing nanoparticle chain models with varying periods or particle sizes, Wood’s anomaly and ultra-narrow absorption cross-sections are observed both before and after H ₂ absorption. Furthermore, the difference in the absorption cross-sections still exhibits Wood's anomaly and ultra-narrow absorption cross-sections. These effects are attributed to the long-range interactions between individual and collective interactions within the unit structures, which can be directly predicted by CD theory. The maximum of the difference absorption cross-sections would be reach to 16.7% when the concentrate of H ₂ changes 4%. The occurrence of Wood's anomaly and ultra-narrow absorption peaks in the absorption spectral difference can effectively indicate the presence of hydrogen in the structure, thereby reflecting the characteristics of hydrogen sensing from another perspective.

Cyber–physical–social systems (CPSSs) emerged as a novel paradigm addressing the growing interactions among physical, social, and cyber domains. Many CPSS challenges were mathematically modeled as optimization problems, with a significant portion requiring multiobjective optimization approaches. To address inefficient learning caused by large-scale distributed sensor data in CPSS, a transformer-based anomaly detection framework for end-edge-cloud systems was proposed. For specific anomaly signals, an improved anomaly detection matrix was developed through systematic calculation of attention weights and corresponding scores. A distributed intelligence architecture integrating transformer-based anomaly detection was implemented in edge nodes for fiber-optic sensors. The proposed artificial Intelligence of Things (AIoT) based distributed edge computing model dynamically allocated tasks among idle, partially available, and fully utilized nodes through optimized resource management in task transmission. A high-efficiency allocation strategy was designed to optimize task processing rates and enable edge nodes to dynamically balance task execution and transmission without compromising detection accuracy. Experimental results demonstrated the effectiveness of both the enhanced transformer method and the computational pipeline across end-edge-cloud architectures. Compared with the vision transformer baseline, our method achieved 3.62% higher accuracy and 3.07% improved recall. Computational time was reduced by approximately 50% while memory consumption decreased by 40%. The proposed approach showed significant potential for anomaly detection in AIoT applications.

With the rise of low-altitude economy applications, 6G communication systems place stricter demands on optical fiber amplifiers, requiring wider bandwidth, higher gain, and better spectral uniformity. Backbone networks for low-altitude integrated sensing and communication systems, in particular, call for high-performance amplification to support robust data transmission and reliable sensing. However, traditional multipump Raman fiber amplifiers (RFAs) are no longer adequate for meeting the performance demands of distributed fiber optic sensing networks in such scenarios. To address this problem, this article proposes a machine learning (ML)-enhanced multipump RFA for a high-performance optical backbone in low-altitude sensing and communication. The back propagation neural network (BPNN) is employed to accurately model the nonlinear relationship between pump parameters and amplification performance, facilitating adaptive and fine-grained control over signal gain, which is critical for maintaining stable and efficient data transmission across dynamic and heterogeneous communication scenarios. Moreover, the artificial bee colony (ABC) algorithm is integrated to perform global optimization of pump wavelengths and power configurations, thereby improving overall system bandwidth, gain characteristics, and operational robustness under diverse and unpredictable network conditions. The experimental results demonstrate that the proposed method achieves superior prediction accuracy, enhanced stability, and greater adaptability compared to conventional algorithms.

With the rapid growth of the low-altitude economy, emerging applications such as urban air mobility and smart logistics demand reliable and low-latency beamforming for uncrewed aerial vehicle-to-vehicle (UAV-to-UAV, U2U) communications in millimeter-wave (mmWave) bands under highly dynamic and nonline-of-sight (NLOS) conditions. Traditional beam alignment methods relying on exhaustive search or channel feedback incur heavy training overhead and degraded accuracy in rapidly varying environments. To address these challenges, we propose multimodal fusion network (MPFusionNet), a multimodal perception-enhanced transformer framework for predictive beamforming. Our approach leverages heterogeneous onboard sensing data including global positioning system (GPS), red–green–blue (RGB) cameras, light detection and ranging (LiDAR), and radar altimeters, incorporates a dynamic time warping (DTW)-based alignment mechanism, and embeds geometry-aware priors within a perceiver input–output (PerceiverIO)-based fusion architecture to achieve robust spatiotemporal representation. Experiments on a simulated U2U dataset show that MPFusionNet attains a top-3 beam prediction accuracy of 97.59%, substantially surpassing conventional models. These results demonstrate the effectiveness of multimodal learning in improving robustness and generalization of predictive beamforming for future autonomous aerial communication systems.

Rule mining in knowledge graphs enables interpretable link prediction. However, deep learning-based rule mining methods face significant memory and time challenges for large-scale knowledge graphs, whereas traditional approaches, limited by rigid confidence metrics, incur high computational costs despite sampling techniques. To address these challenges, we propose MPRM, a novel rule mining method that models rule-based inference as a Markov chain and uses an efficient confidence metric derived from aggregated path probabilities, significantly lowering computational demands. Experiments on multiple datasets show that MPRM efficiently mines knowledge graphs with over a million facts, sampling less than 1% of facts on a single CPU in 22 seconds, while preserving interpretability and boosting inference accuracy by up to 11% over baselines.

In recent years, urban floods have occurred frequently in China. Therefore, there is an urgent need to strengthen urban flood resilience. This paper proposed a hybrid multi-criteria group decision-making method to assess urban flood resilience based on heterogeneous data, group decision-making methodologies, the pressure-state–response model, and social–economic–natural complex ecosystem theory (PSR-SENCE model). A qualitative and quantitative indicator system is formulated using the PSR-SENCE model. Additionally, a new weighting method for indicators, called the synthesis weighting-group analytic hierarchy process (SW-GAHP), is proposed by considering both intrapersonal consistency and interpersonal consistency of decision-makers. Furthermore, an extensional group decision-making technology (EGDMT) based on heterogeneous data is proposed to evaluate qualitative indicators. The flexible parameterized mapping function (FPMF) is introduced for the evaluation of quantitative indicators. The normal cloud model is employed to handle various uncertainties associated with heterogeneous data. The evaluations for Beijing from 2017 to 2021 reveal a consistent annual improvement in urban flood resilience, with a 14.1% increase. Subsequently, optimization recommendations are presented not only for favorable indicators such as regional economic status, drainability, and public transportation service capacity but also for unfavorable indicators like flood risk and population density. This provides a theoretical foundation and a guide for making decisions about the improvement of urban flood resilience. Finally, our proposed method shows superiority and robustness through comparative and sensitivity analyses.