Learning Human-like Knowledge by Singular Value Decomposition: A Progress Report