|Stochastic Systems Group|
This talk will focus on my Masters thesis work, a novel dynamical system model in which the relationship and uncertainty of the future is captured by a nonparametric density estimate. Because of the rapid intractability of such an approach as the length of past dependence grows (due to the increase in dimensionality), we train functionals to reduce the size of the space we must estimate over. This training is performed using a (nonparametric) estimate of mutual information so as to minimize uncertainty of the future for a given level of acceptable computational burden. We apply this approach to several data sets to demonstrate some of its features, including learning a dynamical model for online signatures given relatively few example signatures and admitting a test for autheticity without examples of forgery attempts.
Problems with this site should be emailed to email@example.com