Modeling Natural Image Covariance with a Spike and Slab Restricted Boltzmann Machine

In this work we introduce the spike and slab RBM. The model is characterized by having both a real valued vector: the slab, and a binary variable: the spike, associated with each unit in the hidden layer. The spike and slab RBM possesses some practical properties such as being amenable to Block Gibbs sampling as well as being capable of generating similar latent representations of the data to the recently introduced mcRBM. We also use the spike and slab RBM to argue that in the context of latent variable models, the emphasis placed on explicit modeling of covariance versus explicit models of the predicted mean may be somewhat misplaced.