Using Deep Belief Nets to Learn Covariance Kernels for Gaussian Processes