Distributed Change Detection in Streaming Graph Signals

TL;DR

This paper presents a fully distributed online change-point detection algorithm for streaming graph signals, capable of identifying localized anomalies in large-scale dynamic networks with controlled error rates.

Contribution

It introduces a novel distributed detection method specifically designed for localized anomalies in streaming graph data, with theoretical analysis and simulation validation.

Findings

Effective detection of localized anomalies in simulated networks

Controlled probability of false alarms through analysis

Demonstrated scalability in large dynamic networks

Abstract

Detecting abrupt changes in streaming graph signals is relevant in a variety of applications ranging from energy and water supplies, to environmental monitoring. In this paper, we address this problem when anomalies activate localized groups of nodes in a network. We introduce an online change-point detection algorithm, which is fully distributed across nodes to monitor large-scale dynamic networks. We analyze the detection statistics for controlling the probability of a global type 1 error. Finally we illustrate the detection and localization performance with simulated data.