Jiancheng Jiang (Chair)

Abstract

Independent Component Analysis (ICA) attempts to recover signals from a linear mixture of independent sources. Many existing algorithms have been implemented by modeling the marginal density functions of each source or latent random variable. Motivated by the spatial-temporal nature found in many applications, we approach the ICA problem by treating the sources as independent stochastic processes and develop several algorithms to estimate their features and the mixture parameters.  The talk will highlight methods for modeling the color sources parametrically or non-parametrically. We will describe J. Fan's contribution to one of the areas in estimating the spectral density associated with the latent stochastic processes.