SVD (singular value decomposition)

Also written: SVD, singular value decomposition, np.linalg.svd

Any matrix written as a sum of rank-1 layers, A = U Σ Vᵀ, sorted by singular value. Keeping only the top layers compresses the matrix.