Challenging the Counterintuitive: Revisiting Simple Likelihood Tests with Normalizing Flows for Tabular Data Anomaly Detection

Donghwan Kim, Jung Hun Phee, Hyunsoo Yoon

Keywords: anomaly detection, tabular data, self-supervised learning, generative model

Abstract: In this study, we propose a novel approach to anomaly detection in the tabular domain using normalizing flows, leveraging a simple likelihood test to achieve state-of-the-art performance in unsupervised learning. Although simple likelihood tests have been shown to fail in anomaly detection for image data, we redefine the counterintuitive phenomenon and demonstrate, both theoretically and empirically, why this method succeeds in the tabular domain. Our approach outperforms traditional anomaly detection methods by offering more consistent results. Furthermore, we question the practice of fine-tuning parameters for each dataset individually, ensuring fair and unbiased comparisons by adopting uniform hyperparameters across all datasets. Through extensive experimentation, we validate the robustness and scalability of our method, highlighting its practical effectiveness in real-world settings.

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 9404