Document worth reading: “A Comprehensive Survey on Graph Neural Networks”

Deep learning has revolutionized many machine learning duties these days, ranging from image classification and video processing to speech recognition and pure language understanding. The information in these duties are typically represented throughout the Euclidean home. However, there’s an rising number of functions the place information are generated from non-Euclidean domains and are represented as graphs with superior relationships and interdependency between objects. The complexity of graph information has imposed very important challenges on current machine learning algorithms. Recently, many analysis on extending deep learning approaches for graph information have emerged. In this survey, we provide an entire overview of graph neural networks (GNNs) in information mining and machine learning fields. We counsel a model new taxonomy to divide the state-of-the-art graph neural networks into fully completely different lessons. With a highlight on graph convolutional networks, we overview numerous architectures which have these days been developed; these learning paradigms embody graph consideration networks, graph autoencoders, graph generative networks, and graph spatial-temporal networks. We further focus on the needs of graph neural networks all through various domains and summarize the open provide codes and benchmarks of the current algorithms on fully completely different learning duties. Finally, we propose potential evaluation directions on this fast-growing topic. A Comprehensive Survey on Graph Neural Networks