SuperGaussian Mixture Model UCSD DSP LAB（SuperGaussian混合模型UCSD DSP实验室）.pdfVIP

Journal of Machine Learning Research 0 (0000) 0-0 Submitted 0/00; Published 0/00 Super-Gaussian Mixture Model Jason Palmer Ken Kreutz-Delgado Scott Makeig Department of Electrical and Computer Engineering University of California San Diego La Jolla, CA 90234-0409, USA Editor: XXXX Abstract We propose an extension of the mixture of factor (or independent component) analyzers model to include strongly super-gaussian mixture source densities. This allows greater economy in representation of densities with (multiple) peaked modes or heavy tails than using several Gaussians to represent these features. We derive an EM algorithm to find the maximum likelihood estimate of the model, and show that it converges globally to a local optimum of the actual non-gaussian mixture model without requiring any approximations. This extends considerably the class of source densities that can be used in exact estimation, and shows that strongly super-gaussian densities are as natural as Gaussian densities for these models. We also derive an adaptive Generalized Gaussian algorithm that learns the shape parameters of Generalized Gaussian mixture components. Experiments verify the validity of the algorithm. Keywords: EM, mixture model, non-gaussian 1. Introduction We propose an extension of the mixture of factor (Attias, 1999), or independent component (Choudrey and Roberts, 2002) analyzers model that enlarges the flexibility of the source density mixture model