A Generative Bayesian Graph Attention Network for Semi-Supervised Classification on Scarce Data

Abstract

This research focuses on semi-supervised classification tasks, specifically for graph-structured data under datascarce situations. It is known that the performance of conventional supervised graph convolutional models is mediocre at classification tasks, when only a small fraction of the labeled nodes are given. Additionally, most existing graph neural network models often ignore the noise in graph generation and consider all the relations between objects as genuine ground-truth. Hence, the missing edges may not be considered, while other spurious edges are included. Addressing those challenges, we propose a Bayesian Graph Attention model which utilizes a generative model to randomly generate the observed graph. The method infers the joint posterior distribution of node labels and graph structure, by combining the Mixed-Membership Stochastic Block Model with the Graph Attention Model. We adopt a variety of approximation methods to estimate the Bayesian posterior distribution of the missing labels. The proposed method is comprehensively evaluated on three graph-based deep learning benchmark data sets. The experimental results demonstrate a competitive performance of our proposed model BGAT against the current state of the art models when there are few labels available (the highest improvement is 5%), for semi-supervised node classification tasks.

A Graph Attention Network Based System for Robust Analog Circuits’ Structure Recognition Involving a Novel Data Augmentation Technique

“…GATs can be trained using standard neural network techniques, such as backpropagation. GATs have been shown to outperform traditional graph neural network architectures on a variety of tasks, such as node classification [33] and link prediction [34], [35]. Additionally, GATs are computationally efficient, making them well-suited for large-scale graph data.…”

Contrastive Learning with Heterogeneous Graph Attention Networks on Short Text Classification

“…At the same time, we use the MLP-based model to learn more graph structure information, without an explicit message-passing function. To be more specific, the k-hop neighbours are considered more similar to the target node, where this k th power of the neighbouring information is in the range of [1,2,3,4,5,6,7]. If the neighbouring node is not a k-hop of the target node, the neighbours' information is considered zero.…”

“…On the one hand, most existing studies of GNNs on text classification tasks are trained in a semi-supervised manner, the same as the vanilla Graph Convolution Network (GCN) [5] requiring a large set of labelled data, which cannot be obtained in many real-life scenarios. Therefore, the shortage of labelled data may undermine the performances of graph neural network models on classification tasks, particularly with large scale data [6], [7]. On the other hand, although a GCN can encode local topological properties, it may fail to fully capture the global structural information [8].…”

Is Unimodal Bias Always Bad for Visual Question Answering? A Medical Domain Study with Dynamic Attention

“…For instance, a state-of-the-art model, Vision-Transformer [31], can be deployed for image feature extraction, to improve the current results. Additionally, as there is a limited amount of training data available in medical VQA, we can apply graph generative methods [32], to enhance the generalisation ability of models. • Graph representation learning methods can be introduced to the question embeddings, such as heterogeneous graph neural networks for different words [33], [34], [35].…”