A New Training Algorithm for Kanerva's Sparse Distributed Memory