|
Abstract:
|
Network models for analyzing high-dimensional data are increasingly being used where implicit networks are constructed using measures such as Pearson's correlation and mutual information. When a dependent variable of interest is available, it is useful to construct weighted networks that explicitly take into account the association between the dependent variable and potentially a high-dimensional feature vector. In this presentation, we describe a new method of construction of implicit network using high-dimensional regression. We describe various network wide metrics (NWM) on the network and study the asymptotic properties of NWM when the number of nodes, features, and the sample size increase. We then use these asymptotic properties to identify communities within the implicit network. Extensions of these methods to high-dimensional time series will also be provided.
|
