Attention Approximates Sparse Distributed Memory