|
Abstract:
|
Network analysis is a popular subfield of statistics. The study of a single network is fairly established, but technological advances are making the study of multiple networks increasingly possible for exploring such topics as brain connectomics, molecular representations, and gene co-expression networks. In this presentation, we formulate a Gaussian Process classifier with network inputs. In doing so, we address the challenge of obtaining a provably positive definite kernel by finding a class of distances that yields a PD squared exponential kernel, which also gives rise to a posterior consistency result. Moreover, we outline three computational approaches to our classifier. We present results on both a simulation study and a real dataset. Finally, we discuss how our work applies to the solutions to other statistical problems of interest, including survival analysis, anomaly detection, and hypothesis testing.
|
