My major research interest focuses on designing supervised learning classification algorithms in dynamic graphs. Graph classification is an important graph data mining task that aims to learn a discriminative model from training examples and then use the model to predict class labels of testing examples. Conventional graph classification algorithms (e.g., graph kernels) are designed based on batch mode in which all examples are accessible for training purpose. They will suffer from memory and time issues when applied to dynamic graphs since multiple-scans and holding all data in memory are not realistic in this scenario.

Fundamentally, those restrictions imply the use of incremental strategy. My aim is to develop incremental algorithmic techniques for performing graph classification tasks in dynamic graphs. By retaining important examples from past data and discarding all other examples, it is possible to compress historic data into a feasible size and combine them with current training data to construct a discriminative classifier.