Bankruptcy prediction has long been a significant issue in finance and management science, which attracts the attention of researchers practitioners. With great development modern information technology, it evolved into using machine learning or deep algorithms to do prediction, from initial analysis financial statements. In this paper, we will review models used bankruptcy including classical such as Multivariant Discriminant Analysis (MDA), Logistic Regression (LR), Ensemble method, Neural...

Spatio-temporal neural networks have been successfully applied to weather forecasting tasks recently. The key notion is learn spatio-temporal features concurrently from spatial and temporal dependencies. Existing methods are mainly based on local smoothness assumptions where the learned by accumulating information in regions. However, conditions a certain region usually influenced global meteorological changes long-range historical conditions. Therefore, these that ignore large-scale effects...

Spatiotemporal graph neural networks (GNNs) have been used successfully in traffic prediction recent years, primarily owing to their ability model complex spatiotemporal dependencies within irregular networks. However, the feature extraction processes these methods are limited exploration of inner properties data. Specifically, and temporal convolutions local operations can hardly utilize information from wider ranges, which may affect long-term performance such methods. Furthermore, deep...

Learning brain effective connectivity networks (ECN) from functional magnetic resonance imaging (fMRI) data has gained much attention in recent years. With the successful applications of deep learning numerous fields, several ECN methods based on have been reported literature. However, current ignore temporal features fMRI and fail to fully employ spatial topological relationship between regions. In this article, we propose a novel method for spatiotemporal graph convolutional models...

Advanced graph neural networks have shown great potentials in classification tasks recently. Different from node where embeddings aggregated local neighbors can be directly used to learn labels, requires a hierarchical accumulation of different levels topological information generate discriminative embeddings. Still, how fully explore structures and formulate an effective pipeline remains rudimentary. In this paper, we propose novel network based on supervised contrastive learning with...

Social networks have become indispensable in people's lives. Despite the conveniences brought by social networks, fake news on those online platforms also induces negative impacts and losses for users. With development of deep learning technologies, detecting a data-driven manner has attracted great attention. In this paper, we give brief survey that discusses recent methods detection. Compared with previous surveys, focus different data structures instead models they used to process data....

Recent interests in graph neural networks (GNNs) have received increasing concerns due to their superior ability the network embedding field. The GNNs typically follow a message passing scheme and represent nodes by aggregating features from neighbors. However, current aggregation methods assume that structure is static define local receptive fields under visible connections, which consequently fails consider latent or high-order structures. Besides, are known depth dilemma over-smoothness...

Due to the extraordinary abilities in extracting complex patterns, graph neural networks (GNNs) have demonstrated strong performances and received increasing attention recent years. Despite their prominent achievements, GNNs do not pay enough discriminate nodes when determining information sources. Some of them select sources from all or part neighbors without distinction, others merely distinguish according either structures node features. To solve this problem, we propose concept Influence...

Unsupervised feature learning via auto-encoders results in low-dimensional representations latent space that capture the patterns of input data. The with robust regularization learn qualified features are less sensitive to small perturbations inputs. However, previous highly depend on pre-defined structure settings and often full-connected networks easily prone over-fitting. To solve above limitations, we propose this paper an explicitly regularized framework which improves sparsity...

Discrete network embedding emerged recently as a new direction of representation learning. Compared with traditional models, discrete aims to compress model size and accelerate inference by learning set short binary codes for vertices. However, existing methods usually assume that the structures (e.g., edge weights) are readily available. In real-world scenarios such social networks, sometimes it is impossible collect explicit structure information needs be inferred from implicit data...

Network embedding is a promising topic that maps the vertices to latent space while keeps structural proximity in original space. The network task difficult since have no specific time or orders. Models used extract information from images and texts with regular structures can not be directly applied heading. key feature of methods should further exploited. Previous reviews mainly focus on models algorithms different methods. In this survey, we review works stochastic perspective either data...