This class implements the maximum entropy discrimination (MED) mixture of Gaussians classifier. The discriminant is based on the log ratio of two Gaussian mixture models. The technique permits each Gaussian component to reside in a separate feature space, thus allowing a combination of kernels.