The Nystrom Extension for Signals Defined on a Graph

Abstract

In this paper we introduce a computationally efficient solution to the problem of graph signal interpolation. Our solution is derived using the Nystrom extension and is due to the properties of the Markov matrix which we use as our graph shift operator, inspired by diffusion maps. We focus on graph signals that are smooth over the graph. This assumption cements the relationship between the graph and the graph signal. We experimentally verify our suggested framework on the MNIST data set of handwritten digits.