Learning Eye Vergence Control from a Distributed Disparity Representation