In recent years, graph structures have been widely used in traffic flow prediction. Methods that construct adaptive graphs for data have been shown to outperform models that rely on a single static graph structure. However, most of the current adaptive graphs are static in nature and are constructed in the training set. Since traffic data often suffers from irregular changes in time series, such model architectures are usually unable to reliably, accurately, and timely learn the complex dependencies in traffic data. This paper proposes a new traffic prediction framework - Progressive Multifaceted Dynamic Graph Convolutional Network. Specifically, a progressive adjacency matrix is constructed by learning the trend similarity between graph nodes, and a set of graphs are constructed by gradually adapting to online input data during training and testing phases. Finally, asymptotically dynamic graph convolution is used to capture the complex spatiotemporal relationships in the asymptotic graph to improve traffic prediction accuracy. We apply the model to 4 different real traffic datasets and achieve the best consistency on all datasets.

In recent years, Contrastive learning has shown great potential in traffic flow prediction tasks. However, existing contrastive learning methods have difficulties in dealing with missing data and noise, and it is difficult to fully capture local and global correlations by relying on a single contrast method. In this paper, a Decreasing Mask Spatio-Temporal Graph Comparison Learning Model (DMSTGCL) is proposed. The model dynamically adjusts the mask ratio through the adaptive mask reduction technique to effectively deal with the problem of missing data and noise. Meanwhile, the projection head is further combined with the TripleAttention mechanism in the spatio-temporal contrast learning process, which overcomes the limitations of a single contrast method and captures the complex relationships in local and global space more effectively. Experiments on three real-world datasets demonstrate that DMSTGCL achieves significantly higher prediction accuracy than existing methods.

Sharpness-aware minimization (SAM) has emerged as a highly effective technique to improve model generalization, but its underlying principles are not fully understood. We investigate m-sharpness, where SAM performance improves monotonically as the micro-batch size for computing perturbations decreases, a phenomenon critical for distributed training yet lacking rigorous explanation. We leverage an extended Stochastic Differential Equation (SDE) framework and analyze stochastic gradient noise (SGN) to characterize the dynamics of SAM variants, including n-SAM and m-SAM. Our analysis reveals that stochastic perturbations induce an implicit variance-based sharpness regularization whose strength increases as m decreases. Motivated by this insight, we propose Reweighted SAM (RW-SAM), which employs sharpness-weighted sampling to mimic the generalization benefits of m-SAM while remaining parallelizable. Comprehensive experiments validate our theory and method.

Traffic flow prediction presents significant challenges due to complex spatio-temporal dependencies. Conventional static road network models fail to adequately capture dynamic traffic patterns and struggle with multi-scale feature extraction, limiting prediction accuracy. To address these problems, we present DMAGCN (Dynamic Multi-scale Adaptive Graph Convolutional Network), an innovative architecture combining MGTCN (Multi-scale Gated Temporal Convolution Network) and ADMGCN (Adaptive Dynamic Multi-Graph Convolutional Network) Modules. MGTCN extracts multi-scale temporal features by combining temporal attention mechanisms with gated convolutional networks, while ADMGCN enhances spatial representation through graph convolutions with spatial attention layers and adaptive adjacency matrices. Comprehensive experimental evaluations conducted on the PEMS04 and PEMS08 datasets demonstrate that DMAGCN consistently outperforms existing state-of-the-art methods in traffic flow prediction tasks.

Image denoising is a fundamental task in computer vision, essential for enhancing visual quality and facilitating downstream applications. Traditional denoising methods often rely on handcrafted priors or statistical assumptions, which limit their effectiveness in real-world scenarios with complex noise patterns. Recent advances in deep learning have led to significant improvements by leveraging data-driven approaches. In this paper, we propose an information-enhanced image denoising framework based on deep convolutional neural networks (CNNs), which incorporates both spatial and frequency-domain features to improve denoising performance. Our method integrates a dual-branch architecture: one branch extracts spatial features using a residual CNN, while the other encodes frequency components through discrete wavelet transform (DWT) and learns texture-aware representations. A feature fusion module adaptively combines multi-scale information, enabling robust noise suppression while preserving fine structural details. Extensive experiments on standard benchmarks such as BSD68, Set12, and Urban100 demonstrate that our approach outperforms state-of-the-art denoising methods in terms of PSNR and SSIM, particularly under challenging noise levels and non-Gaussian conditions. The results confirm the effectiveness of information enhancement in deep learning-based image denoising.

Sharpness-Aware Minimization (SAM) has attracted significant attention for its effectiveness in improving generalization across various tasks. However, its underlying principles remain poorly understood. In this work, we analyze SAM’s training dynamics using the maximum eigenvalue of the Hessian as a measure of sharpness and propose a third-order stochastic differential equation (SDE), which reveals that the dynamics are driven by a complex mixture of second- and third-order terms. We show that alignment between the perturbation vector and the top eigenvector is crucial for SAM’s effectiveness in regularizing sharpness, but find that this alignment is often inadequate in practice, which limits SAM's efficiency. Building on these insights, we introduce Eigen-SAM, an algorithm that explicitly aims to regularize the top Hessian eigenvalue by aligning the perturbation vector with the leading eigenvector. We validate the effectiveness of our theory and the practical advantages of our proposed approach through comprehensive experiments. Code is available at https://github.com/RitianLuo/EigenSAM.