Adversarial Attacks for Drift Detection

TL;DR

This paper analyzes the vulnerabilities of existing concept drift detection methods, demonstrating how adversarial data streams can evade detection, and provides theoretical and empirical insights into these shortcomings.

Contribution

It introduces the concept of drift adversarials, systematically constructs them for common schemes, and highlights their impact on drift detection reliability.

Findings

Common drift detection schemes can be bypassed by adversarial data streams.

Theoretical analysis of all possible adversarials for typical detection methods.

Empirical evaluations confirm the effectiveness of drift adversarials.

Abstract

Concept drift refers to the change of data distributions over time. While drift poses a challenge for learning models, requiring their continual adaption, it is also relevant in system monitoring to detect malfunctions, system failures, and unexpected behavior. In the latter case, the robust and reliable detection of drifts is imperative. This work studies the shortcomings of commonly used drift detection schemes. We show how to construct data streams that are drifting without being detected. We refer to those as drift adversarials. In particular, we compute all possible adversairals for common detection schemes and underpin our theoretical findings with empirical evaluations.