Zero-Shot Anomaly Detection (ZSAD) aims to identify and localize anomalous regions in images of unseen object classes. While recent methods based on vision-language models like CLIP show promise, their performance is constrained by existing prompt engineering strategies. Current approaches, whether relying on single fixed, learnable, or dense dynamic prompts, suffer from a representational bottleneck and are prone to overfitting on auxiliary data, failing to generalize to the complexity and diversity of unseen anomalies. To overcome these limitations, we propose PromptMoE. Our core insight is that robust ZSAD requires a compositional approach to prompt learning. Instead of learning monolithic prompts, PromptMoE learns a pool of expert prompts, which serve as a basis set of composable semantic primitives, and a visually-guided Mixture-of-Experts (MoE) mechanism to dynamically combine them for each instance. Our framework materializes this concept through a Visually-Guided Mixture of Prompt (VGMoP) that employs an image-gated sparse MoE to aggregate diverse normal and abnormal expert state prompts, generating semantically rich textual representations with strong generalization. Extensive experiments across 15 datasets in industrial and medical domains demonstrate the effectiveness and state-of-the-art performance of PromptMoE.

The distributed energy management (DEM) problem of smart grids is devoted to achieving optimal energy dispatch and allocation to ensure social welfare maximization. It can be modelled as a distributed optimization problem with physical constraints, whose solution depends on data sharing between smart devices. The exchange of information creates a potential risk for eavesdroppers to intercept and access private data. To avoid privacy leakages, a privacy-preserving optimization strategy via output masks in a consensus framework is proposed to realize social welfare maximization of energy management, where power demand relies on the parameters of the demand sides. Theoretical analysis discloses that eavesdroppers are unable to accurately infer private information of the generators/demand loads. Furthermore, both the convergence and optimality of the proposed strategy are discussed by means of the matrix perturbation theory and the famous KKT optimality condition. Finally, the effectiveness of the proposed strategy for solving the DEM problem is confirmed by the simulation experiments.

This article addresses the integration of model-driven and data-driven approaches for robust hybrid-driven state estimation under limited data and model uncertainties. An unsupervised hybrid estimation framework, termed adaptive model-driven and data-driven (AMD), is proposed. AMD employs an adaptive cross-coupled prior mechanism within the Bayesian inference paradigm to integrate prior information. A two-stage fusion strategy is introduced: an initial hard fusion of model pseudomeasurements and data-driven priors, followed by an adaptive soft fusion that adjusts model influence based on reconstruction discrepancies, thereby enhancing robustness to imperfect model priors. To capture complex nonlinear transition dynamics, a dynamic bilinear recurrent module has been developed, tailored to the system's underlying behavior. The AMD framework adopts a nonidentical training-testing strategy and an unsupervised hybrid learning objective inspired by the information bottleneck principle, enabling accurate parameter learning without access to ground-truth states. Extensive experiments on multiple nonlinear chaotic systems have demonstrated that AMD consistently achieves competitive or superior estimation accuracy compared to state-of-the-art model-based and hybrid approaches, particularly under underdetermined estimation, model mismatch, and dynamic disturbances. These results demonstrate AMD's capability to effectively leverage limited information through complementary fusion, thereby enhancing both data representation and model robustness. This adaptability positions AMD as a powerful solution for challenging state estimation problems.

Accurately predicting real estate enterprise risk is crucial for the national economy. Although some initial works have been made on this topic such as Z-score, support vector machines, and logistic regression, there remains a gap in comprehensive models that can effectively capture the dynamic risk fluctuations from real estate-specific data. As such, a novel prediction model called HRAGRU is proposed for real estate enterprises to forecast potential risk through multimodal data including news reports, policy updates, and stock information in this paper. We first extract the semantic information from news text by using a BERT model optimized for high-target semantic density. Then we investigate the relationships among various data types through a graph neural network (GNN) model with randomly masked edges or nodes. Finally, we establish an improved gated recurrent unit (GRU) model to capture the interactions between new and historical data. The effectiveness of the proposed HRAGRU model is validated using data from A-share and Hong Kong-listed real estate companies, demonstrating its superior performance in forecasting corporate risk indices. Our sources are released at https://github.com/maxiaoyan290/HRAGRU

Image anomaly detection identifies images that deviate from normal patterns. Knowledge Distillation has been extensively studied in unsupervised anomaly detection, where it is assumed that the student model learns and reconstructs the multi-scale normal representations of the pre-trained teacher model, with the representation discrepancies between the student and teacher identified as anomalies. However, the over-generalization of the student model causes it to reconstruct anomalies, leading to the failure of anomaly detection. To address this issue, we propose a novel Diffusion Distillation method for Anomaly Detection (D2AD), which utilizes a lightweight diffusion pipeline to prevent student from imitating abnormal behavior. Specifically, we use the diffusion process to confuse the abnormal student representation with noisy normal representations and restore them to normal, effectively avoiding the over-generalization issue and enabling the student to efficiently transfer the teacher’s refined knowledge of normality. Moreover, we propose the Anomaly-Guided Diffusion Process (AGDP), which dynamically adjusts the noise intensity based on the likelihood of anomalies to balance anomaly confusion and representation recovery. Extensive experiments on the MVTec AD, VisA, and Real-IAD datasets show significant improvements in anomaly detection and localization, while maintaining high efficiency.

This article is concerned with the unified design problem of active fault diagnosis (AFD) and optimal control (OC) for a class of stochastic systems. In AFD, the input is designed to minimize the misdiagnosis probability, whereas the control performance is optimized by designing the input in OC. By exploiting a finite-horizon constrained optimization formulation, a novel unified design framework with normalization strategies is developed for AFD and OC to balance the diagnosis and control performances. The control task is treated as a constraint of the optimization problems to ensure that the task is fulfilled. The Holder divergence family is proposed to construct a new input design criterion for AFD with multifault modes, whose maximization is proven to minimize an upper bound on the misdiagnosis probability. The input design problem under the constructed design criterion for AFD and the input design problem for OC can be solved to global optimality. Based on this global optimality, normalization strategies are utilized to adjust the magnitudes of the objective functions for AFD and OC into the same scale. The quantitative balance between the diagnosis and control performances can be achieved within the developed unified design framework. The effectiveness and superiority of the proposed approach are demonstrated by simulation results on a four-tank system.

Financial time series prediction is an important and challenging data mining task for quantitative investment. The inherent non-linearity, high noise, and susceptibility to various factors, such as macroeconomic conditions and market sentiment in the stock market, increase the difficulty of prediction. Existing financial industries mainly employ time series models or fundamental analysis methods for prediction. However, these methods fail to effectively capture the complex interrelationships between equity. In recent years, graph neural networks (GNNs), due to their powerful relational modeling capabilities, have been applied to stock prediction. However, with the advances of recent digital power, such as widely-used high-frequency trading techniques, existing graph-based methods still have shortcomings in effectively learning multi-granularity temporal relations as they cannot effectively learn the patterns in different frequencies, e.g., minute-level, daily, weekly, etc. Therefore, in this paper, we propose a multi-granularity graph augmented learning framework for interrelated financial time series forecasting. We first construct a temporal return relationship graph with multi-granularity financial time series, including weekly, daily, and minute-level, to comprehensively capture the dynamic relations of equities, including both medium-term trends and short-term fluctuations. Then, to further augment the node relations, we devise an attentional graph augment module to improve the graph learning with fundamental data, which are jointly optimized in the prediction layer. We conduct extensive empirical studies on multiple datasets from both the Chinese and U.S. stock markets. The results demonstrate that our proposed model consistently outperforms existing baseline methods across four key financial metrics, including ARR, ASR, CR, and IR, thereby validating its effectiveness and superiority. The model has been applied and empirically tested in commercial-grade trading platforms, further demonstrating its efficiency and robustness in real-world trading environments.

Stock price prediction is a challenging problem in the field of finance and receives widespread attention. In recent years, with the rapid development of technologies such as deep learning and graph neural networks, more research methods have begun to focus on exploring the interrelationships between stocks. However, existing methods mostly focus on the short-term dynamic relationships of stocks and directly integrating relationship information with temporal information. They often overlook the complex nonlinear dynamic characteristics and potential higher-order interaction relationships among stocks in the stock market. Therefore, we propose a stock price trend prediction model named LSR-IGRU in this paper, which is based on long short-term stock relationships and an improved GRU input. Firstly, we construct a long short-term relationship matrix between stocks, where secondary industry information is employed for the first time to capture long-term relationships of stocks, and overnight price information is utilized to establish short-term relationships. Next, we improve the inputs of the GRU model at each step, enabling the model to more effectively integrate temporal information and long short-term relationship information, thereby significantly improving the accuracy of predicting stock trend changes. Finally, through extensive experiments on multiple datasets from stock markets in China and the United States, we validate the superiority of the proposed LSR-IGRU model over the current state-of-the-art baseline models. We also apply the proposed model to the algorithmic trading system of a financial company, achieving significantly higher cumulative portfolio returns compared to other baseline methods. Our sources are released at https://github.com/ZP1481616577/Baselines\_LSR-IGRU.

This article is concerned with domain generalization (DG), a practical yet challenging scenario in transfer learning where the target data are not available in advance. The key insight of DG is focused on learning a robust model that can generalize to the unseen domain by leveraging knowledge from the source domain. To this end, we propose a novel algorithm known as support-sample-assisted Adversarial Attacks (SSAA) for DG. In the SSAA algorithm, an attack–defense strategy is deployed to enhance the target model's generalizability and transferability. This strategy includes a nontargeted attack stage, during which attack samples are generated to form pseudotarget domains with near-realistic covariate shifts. Subsequently, in the model defense stage, a biclassifier structure is used to distinguish support samples from the generated attack samples. These support samples form a new decision boundary encompassing all unseen samples, prompting an extension of the existing decision boundary to meet these samples. Experimental results on cross-domain fault diagnosis tasks suggest that SSAA outperforms current state-of-the-art DG methods, indicating a promising avenue for further DG development.

This paper is concerned with the distributed filtering problem over wireless sensor networks for a class of state-saturated systems subject to fading measurements and quantization effects. Each sensor node in the network communicates with its neighbors according to the network topology described by a directed graph. The fading phenomena of measurements are assumed to occur in a random way and the attenuation coefficients of the fading measurements are described by a set of random variables with known stochastic properties. By solving two sets of matrix difference equations, an upper bound for the filtering error covariance is presented. Subsequently, with the topology information of the sensor network, such an upper bound is minimized by properly designing the filter parameters. Moreover, the performance of the proposed filter is investigated through establishing sufficient conditions ensuring that the trace of the upper bound is bounded. The relationship between the filter performance and the mean of attenuation coefficient is also discussed. A numerical simulation is exploited to demonstrate the effectiveness of the proposed filtering method.

In this paper, we investigate the distributed filtering problem over wireless sensor networks (WSNs) with bandwidth and energy constraints. To utilize the limited resources efficiently, a novel event-based mechanism is proposed for the sensor node, such that only selected valuable data are broadcasted to its neighboring sensors via the wireless channel according to whether specific events happen. By resorting to graph theory and utilizing stochastic analysis methods, the filter parameters and the event triggering rules are designed, such that the filtering error converges at an exponential rate in the mean square sense. An adaptive algorithm for determining the triggering threshold is developed, which allows the intelligent sensors to tune the boundary of a local event domain in an online manner, so as to keep the average transmission rate level off a desired value. An illustrative example is given to demonstrate the effectiveness of the proposed strategy.