Abstract:Many kernel methods suffer from high time and space complexities and are thus prohibitive in big-data applications. To tackle the computational challenge, the Nystrom method has been extensively used to reduce time and space complexities by sacrificing some accuracy. The Nystrom method speedups computation by constructing an approximation of the kernel matrix using only a few columns of the matrix. Recently, a variant of the Nystrom method called the modified Nystrom method has demonstrated significant improvement over the standard Nystrom method in approximation accuracy, both theoretically and empirically. In this paper, we propose two algorithms that make the modified Nystrom method practical. First, we devise a simple column selection algorithm with a provable error bound. Our algorithm is more efficient and easier to implement than and nearly as accurate as the state-of-the-art algorithm. Second, with the selected columns at hand, we propose an algorithm that computes the approximation in lower time complexity than the approach in the previous work. Furthermore, we prove that the modified Nystrom method is exact under certain conditions, and we establish a lower error bound for the modified Nystrom method.

Convergence Analysis of a Modified Weierstrass Method for the Simultaneous Determination of Polynomial Zeros

On the local convergence of Ehrlich method for numerical computation of polynomial zeros

Chebyshev's Method for Multiple Zeros of Analytic Functions: Convergence, Dynamics and Real-World Applications

A Unified Convergence Theory for Newton-Type Methods for Zeros of Nonlinear Operators in Banach Spaces

Analysis of the Convergence of a Modified Weierstrass Iteration

Convergence of the Modified Halley’s Method for Multiple Zeros under Hölder Continuous Derivative

An Uni ed Convergence Theory for Newton's Type Methods for Zeros of Nonlinear Operators in Banach spaces

ON THE CONVERGENCE OF A MODIFIED CHEBYSHEV-LIKE'S METHOD FOR SOLVING NONLINEAR EQUATIONS

On a Modified Numerov's Method for Inverse Sturm-Liouville Problems

Convergence of the Newton Method and Uniqueness of Zeros of Vector Fields on Riemannian Manifolds

Convergence Analysis of Newton-Like Methods for Inverse Eigenvalue Problems with Multiple Eigenvalues

General convergence theorems for iterative processes and applications to the Weierstrass root-finding method

New simultaneous methods for finding all zeros of a polynomial

Convergence Behavior of Gauss-Newton's Method and Extensions of the Smale Point Estimate Theory.

Convergence of Numerov'S Method for Inverse Sturm-Liouville Problems

Modified Gauss–Seidel type methods and Jacobi type methods for Z-matrices

Efficient Algorithms and Error Analysis for the Modified Nystrom Method.

An effective Linear Independence conjecture for the zeros of the Riemann zeta function and applications

A semi-local convergence theorem for a robust revised Newton's method

On convergence of nourein iterations for simultaneous finding all zeros of a polynomial

Modified Neumann-Neumann methods for semi- and quasilinear elliptic equations