Dynamic and Large-scale Network Survival Analysis

Summary.In this proposal we will study dynamic and large-scale network event data modeling by improvising a unified network survival analysis framework and proposing new modeling strategies. Many complex systems in science and nature consist of interacting individual components which can be represented as nodes with connecting edges in a network and it is becoming increasingly ubiquitous to conduct statistical network analysis in a broad range of scientific disciplines. Examples include social networks (friendship between Facebook users, blog following, twitter following etc.), biological networks (gene network, gene-protein network) and information network (e- mail network, World Wide Web) and many others. Network modeling strategy is one of the brightest gems in modern statistics and provides a powerful and flexible tool for examining the joint occurrence of random interactions among the subjects. It is important to study network survival analysis and develop new techniques to cater for the emerging needs in modeling dynamic and large-scale network event data. Currently there exists limited work that can encompass the temporal dynamics, homophily, degree heterogeneity and varying degree of sparsity for network event data. We have made some initial but important progress on the problem and hope to build on this success to make further study on these new techniques. Long-term impact.The proposal falls within major and active interdisciplinary re- search areas which have attracted many top researchers in statistics, computer science and genetics. Progress on these problems would be of significant interest to researchers in the fields concerned, and the scientific community at large. The research outcomes may have broad implications to the advances of technology in analyzing complex systems to the benefit of the society. The proposed research would be very attractive to graduate students and some of it may also be accessible to undergraduate students. Besides its educational ingredients, the proposed research is also expected to have a broader impact on public policy due to its practical relevance and interdisciplinary nature. By showing that real world network event data can be better modeled by dynamic strategies well accommodating their dynamic characteristics, advanced statistical methodologies and algorithms can well lend support to meaningful and informative conclusions pertaining to protocol creation and public policy making.