Timely anomaly detection in Industrial Internet of Things (IIoT) monitoring requires robust modeling of noisy multivariate sensor streams. Although decomposition-derived residuals provide a potentially useful auxiliary view for multivariate time-series anomaly detection (MTSAD), they are not clean anomaly surrogates in unsupervised settings, as they also contain normal fluctuations, sensor noise, and decomposition artifacts. Therefore, the key challenge is not simply whether residual evidence is useful, but how it can be integrated stably without disturbing the backbone’s native anomaly discrimination process. To address this issue, we propose stable implicit conditioning (SIC), a lightweight and general framework for decomposition-aware unsupervised MTSAD. Instead of reusing residual evidence through explicit downstream intervention, SIC converts compact residual statistics into a bounded sample-dependent channelwise calibration signal for mild representation-level adaptation. When instantiated on the anomaly Transformer (AT), the resulting AT-SIC improves upon the reproduced backbone on four out of five public benchmarks. Additional analyses on decomposition choices, structured disturbances, boundary cases, and cross-backbone transfer show that SIC provides a more reliable evidence utilization path than several intuitive explicit reuse strategies, while introducing only negligible efficiency overhead. These results suggest that decomposition-derived residual evidence is better exploited implicitly than explicitly in this setting.
Guangxia Xu, Zhuo Ye, Xing Huang et al.· IEEE Internet of Things Jour...· 0 citations
Artificial Intelligence-Generated Content (AIGC) has developed rapidly, with Diffusion Models (DMs) gaining wide attention for their superior image generation capabilities. However, the high computational cost of inference limits their deployment in resource-constrained environments such as edge computing. Cloud-edge collaboration is considered a feasible solution to improve inference efficiency, but existing studies have not fully addressed the issue of trust propagation among multiple entities. To address these, we propose a blockchain-aided trusted inference framework for cloud–edge collaborative DMs. Smart contracts are designed to automate image generation task management and ensure traceability throughout the inference process. Leveraging the step-by-step denoising nature of DMs, we introduce a Siamese Network-based Semantic Matching (SNSM) model to identify whether a new task can reuse intermediate results from historical inferences, thereby reducing redundant computation and improving efficiency. We formulate an objective function to minimize total inference latency by jointly considering queuing, transmission, and computation delays, with image quality metrics as constraints. To solve these, we propose Diffusion-Attention integrated Multi-Agent Reinforcement Learning (DAMARL), which dynamically optimizes task partitioning and scheduling between cloud and edge to reduce latency while preserving generation quality. Extensive experiments show that SNSM achieves 85.3% accuracy in reuse decisions, and DAMARL improves average reward by 17.5% $\sim ~35$ % over existing methods, demonstrating the effectiveness of our approach in enhancing DM inference efficiency and performance.
Yu Song, Yinlin Ren, Shaoyong Guo et al.· IEEE Transactions on Cogniti...· 0 citations
Low earth orbit (LEO) satellite networks have emerged as a key enabler for delivering real-time and global services to distributed terrestrial nodes, particularly in remote regions. To preserve data privacy, federated learning (FL) provides a decentralized framework for advancing artificial intelligence (AI) in complex tasks. However, the efficiency of FL is constrained by high and imbalanced energy consumption, which limits its practical deployment. To address these challenges, an energy-aware FL framework that integrates knowledge distillation (KD) with task offloading is proposed, where KD is performed at both the FL server and client devices or direct-connected satellites using public datasets. The energy consumption balancing problem is formulated as a quadratic unconstrained binary optimization (QUBO) model. To achieve computational efficiency and parallelism, the quantum approximate optimization algorithm (QAOA) is employed to solve the problem with both the mixing and cost Hamiltonians derived and the corresponding quantum circuit designed. In a FL framework over a LEO satellite network comprising 40 satellites and 10 FL clients, the proposed method reduces energy consumption by approximately 26.4%, achieves improved energy balance with a weighted variance of approximately 4.93 and maintains high accuracy of 0.95 in a vehicle classification task, compared with the traditional FL method.
Pengxiang Qin, Dongyang Xu, Lei Liu et al.· IEEE Transactions on Cogniti...· 0 citations