Faculty Contact: Ambuj Singh
Abstract: Latent space models are widely used for representing relations among entities with uniquely identifiable and globally significant attributes. This module will focus on understanding how the efficiency of linear processes, including consensus and random walks, performed on graphs embedded in such spaces is affected under the expansion of graph components. In particular, this module will focus on graphs whose edge sets are constructed solely from the interactions of the latent variables associated with the nodes (e.g. geometric graphs). A general problem of interest is improving the speed of convergence for these processes. Goals of this module are to create algorithms which can exactly or approximately select a node from a candidate set which maximizes these desired speed-ups.
Spring 2016: Alex Jones