Tabular anomaly detection fuses foundation-model semantics with domain features
Original title: Universal Semantic Enhancement for Weakly Supervised Anomaly Detection
Authors report the method beats competitors when training uses only eight labeled anomalies.
Researchers propose a weakly supervised anomaly detector for tabular data that combines a pre-trained tabular foundation model with features learned on the target dataset. A prototype-conditioned modulation step integrates the two representations so they reinforce each other. The paper says the approach outperforms competing methods on public datasets when only eight labeled anomalies are available.
Unconfirmed: Outperformance is reported by the authors and is not independently confirmed here.
- The setting assumes abundant unlabeled samples and very few labeled anomalies.
- Universal semantics come from a pre-trained tabular foundation model.
- Domain features capture normal and anomalous patterns in the target data.
- Authors report gains when only eight labeled anomalies are used.
From the original article
1 min read at arxiv.orgWeakly supervised anomaly detection on tabular data is an essential task in many label-scarce applications, since tabular data is widely encountered and it is commonly observed in practice that only a limited number of labeled anomalies are available alongside abundant unlabeled samples. Most existing methods try to learn features that capture the inherent distribution patterns in the target domain.
By Siyu Zhou · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.