About this Research Topic
A critical contribution of modern artificial intelligence (AI) focuses on perception tasks such as pattern recognition and predictive analytics. A key tool for achieving these is representation learning. In the last two decades, graph kernel methods have proved to be one of the most effective methods for graph classification tasks, ranging from the application of disease and brain analysis, chemical analysis, image action recognition and scene modeling, to malware analysis. Even more so, during the last decade, representation learning techniques such as deep neural networks and metric learning on graphs have stimulated fast-increasing attention in light of expanding AI’s success in Euclidean data and sequence data such as images and text. Representation learning on graphs is fundamental to downstream tasks including graph classification, node classification, and graph generation. However, graph-structured data requires to jointly consider graph topology and node/edge attributes, which calls for significant innovations of representation learning techniques for Euclidean and sequential data beyond existing ones.
The inherent challenges in graph-structured data and recent progress of artificial intelligence theories and applications have jointly driven studies in developing advanced representation learning techniques on graph mining and generation. New techniques on graph-structured data have achieved promising performance, particularly when they are applied to applications involving social networks, bio-structural data, and cyber-networks. Nevertheless, there are still numerous challenges regarding the development of expressive and scalable graph embedding and graph generation techniques, especially when graphs become complex, large, dynamic, and involve spatial-temporal characteristics. Finally, it is still unclear whether deep graph learning techniques consistently beat the long-standing graph kernel for graph classification. Consequently, better theoretical understandings between graph neural networks and graph kernel methods are demanded in order to advance and evolve both techniques.
In this Research Topic we invite contributions that introduce new fundamental methods for graph mining and generation that help to develop a better understanding of the reprehensive power of graph neural networks and graph kernels or demonstrate a novel application of these methods to a particular domain. Potential topics include, but are not restricted to:
• Graph embedding
• Deep learning on Graphs
• Scalable methods for large graphs
• Graph kernel methods for graph classification
• Theoretical understanding of graph kernel and graph neural networks
• Metric learning for graph data
• Deep generative models for graph generation/semantic-preserving transformation
• Spatial and temporal graph prediction and generation
• Graph learning in drug discovery
• Graph learning in protein generation
• Graph learning in protein structure prediction
Topic editor Dr. Lingfei Wu is employed by the IBM AI Foundations Lab - Reasoning. All other topic editors declare no competing interests with regards to the Research Topic subject.
Keywords: representation learning, graph mining, graph generation, deep neural networks, graph embedding
Important Note: All contributions to this Research Topic must be within the scope of the section and journal to which they are submitted, as defined in their mission statements. Frontiers reserves the right to guide an out-of-scope manuscript to a more suitable section or journal at any stage of peer review.