As network and autonomous driving technologies rapidly advance, traffic flow prediction has become a crucial area of research. It plays a significant role in optimizing urban traffic management and enhancing road safety, drawing increasing attention from researchers. Traffic flow data, as a specific form of time-series data, are often used in prediction tasks utilizing large language models. Recent developments in graph data and improvements in graph neural networks have led researchers to employ methods like adjacency and Laplacian matrices to address relational issues among distant nodes. However, most existing methods focus on enhancing prediction performance through network architectures or using adaptive matrices to capture spatiotemporal relationships, with limited exploration into the impact of input embedding. This paper introduces the ASTRformer, an innovative approach to traffic flow prediction that emphasizes the fusion of spatial and temporal information in historical data through an adaptive spatio-temporal relation learning mechanism. This mechanism integrates feature embedding with adaptive spatial and temporal embeddings. A learnable spatio-temporal fusion network then parameterizes these embeddings, producing the input representations. Subsequently, a transformer model captures these representations to predict future traffic flows. Experimental results on six datasets demonstrate that our method effectively captures spatiotemporal dependencies, achieving state-of-the-art performance across various prediction metrics.

In unsupervised graph anomaly detection, existing methods usually focus on detecting outliers by learning local context information of nodes, while often ignoring the importance of global context. However, global context information can provide more comprehensive relationship information between nodes in the network. By considering the structure of the entire network, detection methods are able to identify potential dependencies and interaction patterns between nodes, which is crucial for anomaly detection. Therefore, we propose an innovative graph anomaly detection framework, termed CoCo (Context Correlation Discrepancy Analysis), which detects anomalies by meticulously evaluating variances in correlations. Specifically, CoCo leverages the strengths of Transformers in sequence processing to effectively capture both global and local contextual features of nodes by aggregating neighbor features at various hops. Subsequently, a correlation analysis module is employed to maximize the correlation between local and global contexts of each normal node. Unseen anomalies are ultimately detected by measuring the discrepancy in the correlation of nodes’ contextual features. Extensive experiments conducted on six datasets with synthetic outliers and five datasets with organic outliers have demonstrated the significant effectiveness of CoCo compared to existing methods.

The Kalman filter performs well in system state estimation by inferring a joint probability distribution over time variables, which has numerous technological applications in time series analysis. However, the complex Kalman filter parameter settings prevent it from optimally estimating the system state, and this suboptimal estimation makes it difficult to effectively distinguish between normal and anomalous system states in anomaly detection. In this paper, we propose a deep embedding-optimized Kalman filter for unsupervised time series anomaly detection, where the system state of a normal time series can be fitted by the embedding-optimized Kalman filter in an unsupervised manner and anomalies can be detected from data points that deviate from the normal system state. Specifically, we use an autoencoder-enhanced Kalman filter to capture the normal pattern of the time series, where the original time series signal is first fed into the Kalman filter, then the autoencoder encodes the filtered signal and reconstructs the original signal, and the learned embedding from encoder is used to sequentially optimize the Kalman filter. Finally, the optimized filter captures the normal pattern of the time series and the reconstruction error from the filtered signal can be measured to detect anomalies. The validity of the method is verified on real-world time series datasets.

Capturing the complex interaction between the node attribute and the network structure is important for attributed network embedding and anomaly detection. However, there are few methods to explicitly model the correlation between these two views of the node attribute and the network structure. In this article, we propose an attributed network anomaly detection (CaCo) method based on the canonical correlation analysis, which assumes that there should be a strong correlation between the attribute and structure features of normal nodes, and a weak correlation one between those abnormal nodes, in the attributed networks. Consequently, a joint learning mechanism is designed in CaCo to explicitly measure the correlation between two views in the latent space. Specifically, the backbone of a weight-sharing graph convolutional network is employed to encode the node feature from two views of attribute and structure in the latent space, respectively. Then, a Kullback–Leibler divergence regularization is used to align the distributions of the two views. Finally, the parameters of CaCo are optimized by maximizing the correlation between attribute and structure features of normal nodes in the training phase, and anomalies can be detected by measuring the correlation between two views in the testing phase. Extensive experiments on six real-world datasets demonstrate the effectiveness of the proposed method compared to the state-of-the-art techniques.

Cyber-Physical System needs high security to ensure the safe operation. Anomaly detection is one of the mainstream security technologies, the core of which is data analysis and learning. Unsupervised Deep-Learning-based Anomaly Detection Methods can be used in the scenarios that collects large amounts of unlabeled data and are more in line with the actual needs of CPS. However, the correlation among data did not attract enough attention to exploring their implicit relationship, and the adaptive training was deficient. Therefore, we propose an Adaptive-Correlation-aware Unsupervised Deep Learning (ACUDL) for anomaly detection in CPS. It constructs a directed graph structure to represent the implicit correlation among data and adaptively updates with dynamic graph; then, designs a dual-autoencoder to extract the original non-correlation, correlation, and reconstruction features, and builds an estimation network using the Gaussian mixture model (GMM) to estimate the anomaly energy. Experimental results on several CPS data scenarios show that ACUDL can be well adapted to many application scenarios with different data characteristics and achieves better overall results than some up-to-date DL-ADMs.