Reproduce “Anomaly Heterogeneity Learning for Open-set Supervised Anomaly Detection” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jiawen Zhu, Choubo Ding, Yu Tian, Guansong Pang
The Anomaly Heterogeneity Learning (AHL) framework enhances Open-set Supervised Anomaly Detection (OSAD) by explicitly modeling diverse anomaly distributions. This approach improves model generalization to unseen anomaly classes, boosting the Area Under ROC Curve (AUC) of existing OSAD methods by up to 9% across various real-world datasets.
Key insights
- Explicitly modeling and simulating heterogeneous anomaly distributions, rather than assuming homogeneity, is crucial for improving the generalization capabilities of OSAD models to unseen anomaly types.
- Training in simulated open-set environments, where query sets contain unseen normal and anomaly patterns, facilitates the development of robust models better equipped for real-world open-set scenarios.
- Dynamically weighting the contributions of individual anomaly distributions based on their estimated generalization error allows a unified model to synthesize richer and more comprehensive abnormality patterns.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.