What is Wrong with One-Class Anomaly Detection?
This addresses a safety-critical issue for real-world machine learning applications where anomaly detection must handle diverse normal data, though it is incremental in improving existing methods.
The paper tackles the problem of one-class anomaly detection failing when normal samples come from diverse semantic classes, and introduces a latent class-condition-based scenario with a confidence-based self-labeling framework that outperforms recent one-class methods in such scenarios.
From a safety perspective, a machine learning method embedded in real-world applications is required to distinguish irregular situations. For this reason, there has been a growing interest in the anomaly detection (AD) task. Since we cannot observe abnormal samples for most of the cases, recent AD methods attempt to formulate it as a task of classifying whether the sample is normal or not. However, they potentially fail when the given normal samples are inherited from diverse semantic labels. To tackle this problem, we introduce a latent class-condition-based AD scenario. In addition, we propose a confidence-based self-labeling AD framework tailored to our proposed scenario. Since our method leverages the hidden class information, it successfully avoids generating the undesirable loose decision region that one-class methods suffer. Our proposed framework outperforms the recent one-class AD methods in the latent multi-class scenarios.