Traffic flow prediction has become an important component of intelligent transportation systems. However, high-precision traffic flow prediction (especially long-term prediction) is still very challenging due to the complex spatial-temporal dependence of urban traffic data. In this paper, a novel Multi-scale Temporal and Enhance Spatial Transformer (MTESformer) model is proposed to capture complex dynamic spatial-temporal dependencies. MTESformer provides a reasonable feature embedding of periodic characteristics of traffic; it can recognize different temporal feature patterns and capture long-term dependencies, and efficiently focuses on two different node-space dependencies (long-range and neighboring nodes dependencies). Specifically, we develop a special multi-scale convolution unit that unites temporal self-attention to capture a wider range of dynamic temporal dependencies from a multi-receptive field and identify different temporal feature patterns. Secondly, we design a novel Enhance Spatial Transformer module, which can better focus on the dynamic spatial dependencies among nodes by fusing their neighborhood information. Experimental results on the public transportation network datasets METR-LA, PEMS-BAY, PEMS04, and PEMS08 data show that our proposed method outperforms most of the baseline models and outperforms the state-of-the-art models in long-term prediction. (The MAE of 60min prediction of our model on METR-LA, PEMS-BAY dataset is 3.37, 1.87, and the MAPE is 9.62%, 4.35%, respectively, and all of them outperform the PDFormer on PEMS04 and PEMS08 dataset).

Semi-supervised learning (SSL) is critical for medical image segmentation but often struggles with network dependency and pseudo-label error accumulation. To address these issues, we propose a fractal-dimension-guided hierarchical contrastive learning dual-student framework(FD-HCL). We extend the Mean Teacher architecture with a dual-student design and introduce an independence-aware exponential moving average (I-EMA) update mechanism to mitigate model coupling. For enhanced feature learning, we devise a hierarchical contrastive learning (HCL) mechanism guided by voxel uncertainty, spanning global, high-confidence, and low-confidence regions. We further improve structural integrity by incorporating a fractal-dimension (FD)-weighted consistency loss and integrating a novel uncertainty-aware bidirectional copy–paste (UB-CP) augmentation. Extensive experiments on the LA and BraTS 2019 datasets demonstrate the state-of-the-art performance of our framework across 10% and 20% labeled data settings. On the LA dataset with 10% labeled data, our method achieved a Dice score that outperformed the best existing approach by 0.68%. Similarly, under the 10% labeling setting on the BraTS 2019 dataset, we surpassed the state-of-the-art Dice score by 0.55%.

Graph anomaly detection (GAD) is crucial for safeguarding the integrity and security of complex systems, such as social networks and financial transactions. Despite the advances made by Graph Neural Networks (GNNs) in the field of GAD, existing methods still exhibit limitations in capturing subtle structural anomaly patterns: they typically over-rely on reconstruction error, struggle to fully exploit structural similarity among nodes, and fail to effectively integrate attribute and structural information. To tackle these challenges, this paper proposes a structure-aware dual-path attention graph node anomaly detection method (DPGAD). DPGAD employs wavelet diffusion to extract network neighborhood features for each node while incorporating a dual attention mechanism to simultaneously capture attribute and structural similarities, thereby obtaining richer feature details. An adaptive gating mechanism is then introduced to dynamically adjust the fusion of attribute features and structural features. This allows the model to focus on the most relevant features for anomaly detection, enhancing its robustness and antinoise capability. Our experimental evaluation across multiple real-world datasets demonstrates that DPGAD consistently surpasses existing methods, achieving average improvements of 9.06% in AUC and 11% in F1-score. Especially in scenarios where structural similarity is crucial, DPGAD has a performance advantage of more than 20% compared with the most advanced methods.

The development of deep learning models for intelligent vehicles rely on a large number of reliable data, among which large-scale and accurately labeled traffic scene image data is conducive to promoting the research of intelligent perception algorithms. In contrast to the time-consuming and laborious process of manual annotation, the virtual dataset synthesis method is a convenient and efficient way to replace manual labeling. However, due to the existence of automatically labeled samples in the virtual dataset that are not conducive to model training, and the domain shift between the virtual dataset and the real scene, the model trained from the virtual dataset cannot be well generalized to the real scene. To address these issues, we propose a two-stage label screening strategy for the data processing to remove the interference samples with low information content and further propose a novel efficient traffic scene dataset named Large-Scale Software-Defined Transportation Virtual Dataset (LSTVD). Moreover, an attention-induced style alignment domain adaptation method (ASADA) is proposed to reduce the distribution difference between the virtual data domain and the real-world data domain. Experiment results demonstrate the efficiency of our label screening strategy and our domain adaptation method.