Academic

ICLAD: In-Context Learning for Unified Tabular Anomaly Detection Across Supervision Regimes

arXiv:2603.19497v1 Announce Type: new Abstract: Anomaly detection on tabular data is commonly studied under three supervision regimes, including one-class settings that assume access to anomaly-free training samples, fully unsupervised settings with unlabeled and potentially contaminated training data, and semi-supervised settings with limited anomaly labels. Existing deep learning approaches typically train dataset-specific models under the assumption of a single supervision regime, which limits their ability to leverage shared structures across anomaly detection tasks and to adapt to different supervision levels. We propose ICLAD, an in-context learning foundation model for tabular anomaly detection that generalizes across both datasets and supervision regimes. ICLAD is trained via meta-learning on synthetic tabular anomaly detection tasks, and at inference time, the model assigns anomaly scores by conditioning on the training set without updating model weights. Comprehensive experi

J
Jack Yi Wei, Narges Armanfard · · 1 min read · 8 views
#cs.LG

arXiv:2603.19497v1 Announce Type: new Abstract: Anomaly detection on tabular data is commonly studied under three supervision regimes, including one-class settings that assume access to anomaly-free training samples, fully unsupervised settings with unlabeled and potentially contaminated training data, and semi-supervised settings with limited anomaly labels. Existing deep learning approaches typically train dataset-specific models under the assumption of a single supervision regime, which limits their ability to leverage shared structures across anomaly detection tasks and to adapt to different supervision levels. We propose ICLAD, an in-context learning foundation model for tabular anomaly detection that generalizes across both datasets and supervision regimes. ICLAD is trained via meta-learning on synthetic tabular anomaly detection tasks, and at inference time, the model assigns anomaly scores by conditioning on the training set without updating model weights. Comprehensive experiments on 57 tabular datasets from ADBench show that our method achieves state-of-the-art performance across three supervision regimes, establishing a unified framework for tabular anomaly detection.

Executive Summary

This study proposes ICLAD, a novel deep learning approach for unified tabular anomaly detection across various supervision regimes. ICLAD is trained via meta-learning on synthetic tasks and leverages in-context learning to assign anomaly scores without updating model weights. Comprehensive experiments on 57 datasets demonstrate its state-of-the-art performance across one-class, fully unsupervised, and semi-supervised settings. This work establishes a unified framework for tabular anomaly detection, enabling the generalization of models across different data distributions and supervision levels. The study's findings have significant implications for various applications, including fraud detection, quality control, and data cleaning.

Key Points

  • ▸ ICLAD is a novel deep learning approach for unified tabular anomaly detection.
  • ▸ ICLAD leverages meta-learning and in-context learning for anomaly score assignment.
  • ▸ Comprehensive experiments demonstrate ICLAD's state-of-the-art performance across three supervision regimes.

Merits

Generalizability

ICLAD's ability to generalize across different data distributions and supervision levels enables its application in various real-world scenarios.

Efficiency

ICLAD's in-context learning mechanism allows for efficient anomaly score assignment without requiring model weight updates.

Scalability

ICLAD's unified framework can be applied to large-scale tabular datasets, making it a valuable tool for data analysis.

Demerits

Complexity

The development and training of ICLAD may be computationally intensive, requiring significant resources and expertise.

Interpretability

The lack of interpretability in ICLAD's anomaly score assignment mechanism may hinder its adoption in high-stakes applications.

Robustness

ICLAD's reliance on meta-learning and in-context learning may compromise its robustness to adversarial attacks or distribution shifts.

Expert Commentary

The study's proposal of ICLAD marks a significant advancement in the field of tabular anomaly detection. However, the study's limitations in terms of complexity, interpretability, and robustness must be addressed to ensure its adoption in high-stakes applications. The study's findings also underscore the importance of explainable AI and transfer learning in deep learning architectures. As such, further research in these areas is warranted to unlock the full potential of ICLAD and related techniques.

Recommendations

  • ✓ Develop and deploy ICLAD in high-stakes applications, such as fraud detection and quality control, to evaluate its practical performance and limitations.
  • ✓ Investigate methods to improve ICLAD's interpretability and robustness, such as incorporating explainable AI techniques and adversarial training.

Sources

Original: arXiv - cs.LG

Related Articles