Research Seminar by Dr. Yuan Fang | Harnessing Graph Data: Principles and Applications

Big data are no doubt complex: there exist not only individual data entities, but also various interactions between one entity and another. Fortunately, a graph is capable of modelling both entities and their interactions as its nodes and edges, respectively. Real-life graphs are abundant, such as the Web, social networks, transportation systems and protein-protein interactions, often entailing scientific, engineering and social significance. Furthermore, traditional flat data can often be transformed into graphs for a richer representation. To gain insights into such data, my research has undertaken data mining and machine learning on graphs. In particular, I have studied fundamental principles of learning on graphs. First, to exploit graph topology, we developed novel random walk models. Second, to understand the behavior of neighboring nodes on graphs, we proposed the principle of pointwise smoothness. Third, to exploit the rich semantics embedded on a heterogeneous graph, we investigated the novel concept of metagraph representations. Together with an efficient and scalable computational framework, these principles can be applied to solve important problems on graphs including ranking and classification, enabling a wide range of user, entity, and knowledge-centric applications.
 

About the Speaker