Attention approximates Sparse Distributed Memory