Anomaly detection on tabular data is a core problem in settings where labels are scarce, according to a new preprint on arXiv. The authors argue that weakly supervised anomaly detection is essential because tabular data is widely encountered and in practice only a limited number of labeled anomalies is usually available.

The paper, titled "Universal Semantic Enhancement for Weakly Supervised Anomaly Detection," proposes a method that applies semantic enhancement to improve performance in this setting. The abstract does not yet provide specific technical details or experimental results.

Since this is the only source and it is a preprint, the claims have not been peer-reviewed. Further information would be needed to evaluate the method's effectiveness.