Reversibility of elliptical slice sampling revisited
We discuss the well-definedness of elliptical slice sampling, a Markov chain approach for approximate sampling of posterior distributions introduced by Murray, Adams and MacKay 2010. We point to a regularity requirement and provide an alternative proof of the reversibility property. In particular, this guarantees the correctness of the slice sampling scheme also on infinite-dimensional separable Hilbert spaces.
READ FULL TEXT