Out-of-Distribution Detection with Reconstruction Error and Typicality-based Penalty

TL;DR

This paper introduces PRE, a novel out-of-distribution detection method combining reconstruction error with a typicality-based penalty using normalizing flows, effectively identifying OOD and adversarial examples in image datasets.

Contribution

It proposes a new OOD detection approach that integrates typicality-based penalties with reconstruction error in normalizing flows, addressing limitations of previous methods.

Findings

PRE outperforms existing OOD detection methods on CIFAR-10, TinyImageNet, and ILSVRC2012 datasets.

The method effectively detects adversarial and OOD examples.

Normalizing flows with the proposed penalty improve detection accuracy.

Abstract

The task of out-of-distribution (OOD) detection is vital to realize safe and reliable operation for real-world applications. After the failure of likelihood-based detection in high dimensions had been shown, approaches based on the \emph{typical set} have been attracting attention; however, they still have not achieved satisfactory performance. Beginning by presenting the failure case of the typicality-based approach, we propose a new reconstruction error-based approach that employs normalizing flow (NF). We further introduce a typicality-based penalty, and by incorporating it into the reconstruction error in NF, we propose a new OOD detection method, penalized reconstruction error (PRE). Because the PRE detects test inputs that lie off the in-distribution manifold, it effectively detects adversarial examples as well as OOD examples. We show the effectiveness of our method through the evaluation using natural image datasets, CIFAR-10, TinyImageNet, and ILSVRC2012.