TaskBridge: Bridging Unsupervised Tabular Anomaly Detection and In-Context Learning via Virtual Tasks

2026-09-29

AI Summary

This study addresses the reliance of existing unsupervised tabular anomaly detection methods on task-specific pretraining and their prohibitive computational costs. To overcome these limitations, this work proposes reformulating anomaly detection as a contextual reasoning task for tabular foundation models. By constructing virtual supervision tasks that induce the model to predict underlying data structures, the approach directly generates anomaly evidence without requiring anomaly-specific pretraining, thereby efficiently repurposing general-purpose tabular foundation models. This method achieves effective zero-shot anomaly detection, significantly outperforming thirty baseline approaches across 790 real-world datasets. Ultimately, it establishes a new paradigm for unsupervised tabular anomaly detection that is both highly generalizable and scalable.

Abstract

Unsupervised tabular anomaly detection (TAD) aims to identify anomalous rows in tabular data using normal training samples. While conventional methods rely on dataset-specific training and configuration search, recent tabular foundation models (TFMs) enable zero-shot anomaly detection on unseen datasets via in-context learning. Most TFM-based approaches, however, require anomaly-specific pretraining from scratch, making detection inherently dependent on synthetic TAD-specific priors and costly to update. Some approaches instead repurpose pretrained general-purpose TFMs for TAD to avoid this burden, but rely on computationally expensive formulations with restrictive anomaly inductive biases. In this work, we introduce TaskBridge, a new framework that efficiently repurposes pretrained general-purpose TFMs for unsupervised TAD by constructing virtual supervised tasks that directly recast anomaly detection as supervised in-context inference of TFMs. The resulting virtual tasks induce predictive structures under which normal queries and their target pairs receive high support, whereas anomalies tend to violate the induced structures and receive lower support, providing direct anomaly evidence. Across 790 real-world datasets, TaskBridge consistently outperforms 30 baselines, including state-of-the-art TFM-based approaches, without anomaly-specific TFM pretraining or dataset-specific model optimization.

Problem

• Unsupervised Tabular Anomaly Detection
• Tabular Foundation Models
• In-Context Learning
• Zero-shot Anomaly Detection

Innovation

• Tabular Anomaly Detection
• In-Context Learning
• Tabular Foundation Models
• Virtual Tasks
• Zero-shot