Non-Robust Features are Not Always Useful in One-Class Classification
This addresses the robustness problem for practitioners using lightweight one-class classifiers in applications like anomaly detection, but it is incremental as it builds on prior work on adversarial examples.
The paper investigates the vulnerability of lightweight one-class classifiers to adversarial attacks, finding that they learn non-robust features like texture, but unlike in multi-class classification, these features are not always useful for the one-class task, indicating an unwanted training consequence.
The robustness of machine learning models has been questioned by the existence of adversarial examples. We examine the threat of adversarial examples in practical applications that require lightweight models for one-class classification. Building on Ilyas et al. (2019), we investigate the vulnerability of lightweight one-class classifiers to adversarial attacks and possible reasons for it. Our results show that lightweight one-class classifiers learn features that are not robust (e.g. texture) under stronger attacks. However, unlike in multi-class classification (Ilyas et al., 2019), these non-robust features are not always useful for the one-class task, suggesting that learning these unpredictive and non-robust features is an unwanted consequence of training.