Object oriented data analysis of networks

Project description

Network data is now routinely available from a variety of applications, including in social media, corpus linguistics and neuroimaging. Less common is the study of samples of networks, for example collected over time or at random from a population. The project will take an object oriented data analysis approach, where the first questions of interest are what are the data objects, what space do they lie and how are they represented in feature space. Networks can be compared by using metrics on the space of graph Laplacians, with the Frobenius norm being used most commonly. We will develop statistical methodology using other metrics, and also develop statistical procedures in the resulting manifolds. Motivating applications include large social and financial datasets from developing economies. Such datasets are often very sparse and very noisy, and so the appropriate handling of uncertainty in the analysis of the networks is paramount.

This project will be jointly supervised by James Goulding (N/LAB, Business School).