We investigate the generalizability of deep learning based on the sensitivity to input perturbation. We hypothesize that the high sensitivity to the perturbation of data degrades the performance on it. To reduce the sensitivity to perturbation, we propose a simple and effective regularization method, referred to as spectral norm regularization, which penalizes the high spectral norm of weight m...

We develop and analyze a first-order system least-squares spectral method for the second-order elliptic boundary value problem with variable coefficients. We first analyze the Chebyshev weighted norm least-squares functional defined by the sum of the Lwand H−1 w norm of the residual equations and then we replace the negative norm by the discrete negative norm and analyze the discrete Chebyshev ...

A function f : F2 → {−1, 1} is called linear-isomorphic to g if f = g ◦ A for some non-singular matrix A. In the g-isomorphism problem, we want a randomized algorithm that distinguishes whether an input function f is linear-isomorphic to g or far from being so. We show that the query complexity to test g-isomorphism is essentially determined by the spectral norm of g. That is, if g is close to ...

The publication of the important work Rauch and Taylor [RT75] started a hole branch research on wild perturbations Laplace-Beltrami operator. Here, we extend certain results show norm convergence resolvent. We consider (not necessarily compact) manifold with many small balls removed, number can increase as radius is shrinking, also be infinite. If distance shrinks less fast than radius, then th...

In this paper, we consider the Legendre spectral Galerkin and Legendre spectral collocation methods to approximate the solution of Urysohn integral equation. We prove that the approximated solutions of the Legendre Galerkin and Legendre collocation methods converge to the exact solution with the same orders, O(n−r) in L2-norm and O(n 1 2 −r) in infinity norm, and the iterated Legendre Galerkin ...