In deep learning the hessian is not feasible to represent.
Are iterative techniques to preform various operations such as approximating matrix inversion, approximating eigenvectors (eigenvalues) using only matrix vector products.
