Bridging the Gap between Spatial and Spectral Domains: A Survey on Graph Neural Networks

The success of deep learning has been widely recognized in many machine learning tasks during the last decades, ranging from image classification and speech recognition to natural language understanding. As an extension of deep learning, Graph neural …

Graph Mining

[ with University of Taxes at Dallas, Rensselaer Polytechnic Institute ] Today’s deep learning is still a black box. Similarly, graph neural network is another black box. This incurs difficult in comparison and improvement since each method is acclaimed as state-of-the-art. A unified framework is needed to avoid the potential risks, which is our goal.

Spatial Temporal Mining

[ with Virginia Tech ] Graph neural networks (GNNs) motivate many applications based on network data such as transportation road. However, as a practical scenario, the road has very different characteristics from a smooth network (e.g., social network) often used by GNN experiments. We plan to identify these special features and accordingly adjust GNN to adopt transportation problems.

Materials Discovery

[ with Toyota Research, Illinois Tech ] Material development heavily relies on domain knowledge and professional’s intuitive, which hinders the discovery of new material. We are collaborating with material experts and propose generative models for boosting material discovery.

Circuit Security and SAT solving

[ with UC Davis, George Manson University ] Boolean Satisfiability (SAT) problem is a fundamental problem in computer science and the core of many real-world applications such as hardware and software design. However, its solving or estimating hardness of SAT is NP-hard. We are seeking for a series of a methodology by deep learning to significantly accelerate its solving and estimation.