Exploiting Restricted Boltzmann Machines and Deep Belief Networks in Compressed Sensing

摘要

This paper proposes a CS scheme that exploits the representational power ofrestricted Boltzmann machines and deep learning architectures to model theprior distribution of the sparsity pattern of signals belonging to the sameclass. The determined probability distribution is then used in a maximum aposteriori (MAP) approach for the reconstruction. The parameters of the priordistribution are learned from training data. The motivation behind thisapproach is to model the higher-order statistical dependencies between thecoefficients of the sparse representation, with the final goal of improving thereconstruction. The performance of the proposed method is validated on theBerkeley Segmentation Dataset and the MNIST Database of handwritten digits.