Skip to content

Introduction to GSKS

Chenhan D. Yu edited this page Jan 8, 2017 · 8 revisions

GSKS stands for General Stride Kernel Summation. It computes kernel matrix-vector product in a matrix-free fashion. That is to say, while computing kernel summations for certain columns and rows, GSKS does not require extra temporary space to collect coordinates and the kernel matrices are not stored. Kernel summation is widely used in kernel methods, density estimation, Gaussian process, computational statistics.

GSKS (General Stride Kernel Summation)

Definition

GSKS< OP1, OP2, OPKERNEL, T > computes the following:

gkmm definition

A and B are coordinates stored in d dimension leading format. Potentials u records the weighted average (with weights w) of a sub kernel matrix created by A and B.

Templates

Algorithms

Limitations