TabADM: Unsupervised Tabular Anomaly Detection with Diffusion Models

TL;DR

TabADM introduces a diffusion-based probabilistic model for unsupervised anomaly detection in tabular data, effectively identifying anomalies without requiring clean datasets or extensive tuning.

Contribution

The paper presents a novel diffusion model that learns normal data density from contaminated datasets using a rejection scheme, enhancing anomaly detection in tabular data.

Findings

Improves detection capabilities over baseline methods.

Stable across different data dimensions.

Requires minimal hyperparameter tuning.

Abstract

Tables are an abundant form of data with use cases across all scientific fields. Real-world datasets often contain anomalous samples that can negatively affect downstream analysis. In this work, we only assume access to contaminated data and present a diffusion-based probabilistic model effective for unsupervised anomaly detection. Our model is trained to learn the density of normal samples by utilizing a unique rejection scheme to attenuate the influence of anomalies on the density estimation. At inference, we identify anomalies as samples in low-density regions. We use real data to demonstrate that our method improves detection capabilities over baselines. Furthermore, our method is relatively stable to the dimension of the data and does not require extensive hyperparameter tuning.