Non-Linear Outlier Synthesis for Out-of-Distribution Detection

Lars Doorenbos

Keywords: Out-of-Distribution Detection, Outlier Synthesis, Diffusion Models

TL;DR: We present NCIS, which can generate informative outliers for out-of-distribution detection by modeling class-conditional manifolds in diffusion embedding space.

Abstract: The reliability of supervised classifiers is severely hampered by their limitations in dealing with unexpected inputs, leading to great interest in out-of-distribution (OOD) detection. Recently, OOD detectors trained on synthetic outliers, especially those generated by large diffusion models, have shown promising results in defining robust OOD decision boundaries. Building on this progress, we present Non-Linear Class-wise Invariant Sampling (NCIS), which enhances the quality of synthetic outliers by operating directly in the diffusion model's embedding space, rather than combining disjoint models as in previous work, and by modeling class-conditional manifolds with a conditional volume-preserving network, allowing for a more expressive characterization of the training distribution. We demonstrate that these improvements yield new state-of-the-art OOD detection results on standard ImageNet100 and CIFAR100 benchmarks and provide insights into the importance of data pre-processing and other key design choices. We will make our code available upon acceptance.

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/2026/AuthorGuide.

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: 8810