Graph Neural Networks (GNNs): Foundation, Frontiers and Applications

Graph structured data such as social networks and molecular graphs are omnipresent in the real world. Developing sophisticated algorithms for representation learning on graph structured data holds significant research value as it enables smoother execution of subsequent tasks. Graph Neural Networks (GNNs) provide a novel approach to learning representations for graph structured data, extending the capabilities of deep neural network models. These GNNs offer effective means to acquire representations at both the node and graph levels. The commendable representation learning ability of GNNs has led to their practical importance in a wide range of applications, including recommendation systems, natural language processing, and healthcare. It has become one of the hottest research topics and attracted significantly increasing attention from the machine learning and data mining communities in recent years. Despite graph neural networks receiving remarkable attention, the challenges still remain when applying them into other domains, from the theoretical understanding of methods to the interpretability in a real industry scenario, and from the reliability of the methodology to its practical performance in a specific application.

Numerous research avenues are currently underway or anticipated for future exploration, each offering valuable opportunities for further investigation. These include the development of novel models tailored to unexplored graph structures, the enhancement of compositional aspects within existing models, addressing dynamic graphs, exploring interpretability and robustness, and more. However, given the rapid expansion of the field, attaining a comprehensive understanding of the advancements in Graph Neural Networks (GNNs) has proven to be exceedingly difficult. Therefore, we feel the urgency to bridge the above gap and have a dedicated collection of journal literatures to summarize these fast growing yet challenging topics in GNNs.

This Research Topic aims to bring together both academic and industrial researchers from different backgrounds and perspectives to above challenges. The interested topics include: 1) a thorough understanding of GNNs theories and basic concepts; 2) use of state-of-the-art algorithms to explore major recent advances in GNNs research; and 3) delve into uncharted research prospects for GNNs and examine the utilization and potential design of GNN algorithms for practical applications in the real world.

The foundation and advanced problems include but not limited to:
• Representation learning on graphs.
• Graph neural networks on node classification, graph classification, link prediction
• The expressive power of Graph neural networks
• Interpretability in Graph Neural Networks
• Graph neural networks for graph matching
• Graph structure learning
• Dynamic/incremental graph-embedding
• Learning representation on heterogeneous networks, knowledge graphs
• Deep generative models for graph generation
• Graph Neural Networks: AutoML
• Graph2seq, graph2tree, and graph2graph models
• Spatial and temporal graph prediction and generation

And with particular focuses but not limited to these applications:
• Natural language processing
• Bioinformatics (drug discovery, protein generation, protein structure prediction)
• Graph Neural Networks Program synthesis and analysis and software mining
• Deep learning in neuroscience (brain network modeling and prediction)
• Cybersecurity (authentication graph, Internet of Things, mal- ware propagation)
• Geographical network modeling and prediction (Transportation and mobility networks, Internet, mobile phone networks, power grids, social and contact networks


Topic Editors Xiaojie Guo and Lingfei Wu are employed by IBM T.J. Watson Research Center and Pinterest Inc. respectively. All other Topic Editors declare no competing interests with regards to the Research Topic subject.