Anomaly Detection based on Compressed Data: an Information Theoretic Characterization

TL;DR

This paper investigates how lossy compression affects anomaly detection in sensor signals, using information theory to formalize the relationship and validating findings with analytical derivations and practical detector performance.

Contribution

It provides an information-theoretic framework for understanding the impact of compression on anomaly detection, including analytical derivations within the Gaussian model.

Findings

Higher compression quality loss reduces detection accuracy

Analytical results align with practical detector performance

Framework applicable in asymptotic signal regimes

Abstract

We analyze the effect of lossy compression in the processing of sensor signals that must be used to detect anomalous events in the system under observation. The intuitive relationship between the quality loss at higher compression and the possibility of telling anomalous behaviours from normal ones is formalized in terms of information-theoretic quantities. Some analytic derivations are made within the Gaussian framework and possibly in the asymptotic regime for what concerns the stretch of signals considered. Analytical conclusions are matched with the performance of practical detectors in a toy case allowing the assessment of different compression/detector configurations.