Adaptive Graph Normalized Sign Algorithm

TL;DR

The paper introduces the Graph Normalized Sign (GNS) algorithm, an adaptive method for graph signal prediction that is robust to impulsive noise and missing data, offering faster convergence and lower error.

Contribution

It proposes the GNS algorithm, combining graph signal processing with a normalized sign approach for improved adaptive prediction.

Findings

GNS achieves faster convergence than previous methods.

GNS provides robust predictions under impulsive noise.

GNS demonstrates lower error in temperature data prediction.

Abstract

Efficient and robust prediction of graph signals is challenging when the signals are under impulsive noise and have missing data. Exploiting graph signal processing (GSP) and leveraging the simplicity of the classical adaptive sign algorithm, we propose an adaptive algorithm on graphs named the Graph Normalized Sign (GNS). GNS approximated a normalization term into the update, therefore achieving faster convergence and lower error compared to previous adaptive GSP algorithms. In the task of the online prediction of multivariate temperature data under impulsive noise, GNS outputs fast and robust predictions.