Neural network learns low-dimensional polynomials with SGD near the information-theoretic limit

Open in new window