2DSig-Detect: a semi-supervised framework for anomaly detection on image data using 2D-signatures

TL;DR

This paper presents 2DSig-Detect, a semi-supervised anomaly detection framework for images that leverages 2D-signatures and rough path theory to improve detection of adversarial attacks with better performance and efficiency.

Contribution

The paper introduces a novel 2D-signature-based semi-supervised framework for image anomaly detection, specifically targeting adversarial attacks, with demonstrated superior performance and reduced computation time.

Findings

Outperforms existing methods in adversarial attack detection

Reduces computation time for identifying adversarial perturbations

Effective in both training-time and test-time attack scenarios

Abstract

The rapid advancement of machine learning technologies raises questions about the security of machine learning models, with respect to both training-time (poisoning) and test-time (evasion, impersonation, and inversion) attacks. Models performing image-related tasks, e.g. detection, and classification, are vulnerable to adversarial attacks that can degrade their performance and produce undesirable outcomes. This paper introduces a novel technique for anomaly detection in images called 2DSig-Detect, which uses a 2D-signature-embedded semi-supervised framework rooted in rough path theory. We demonstrate our method in adversarial settings for training-time and test-time attacks, and benchmark our framework against other state of the art methods. Using 2DSig-Detect for anomaly detection, we show both superior performance and a reduction in the computation time to detect the presence of adversarial perturbations in images.