Positive Difference Distribution for Image Outlier Detection using Normalizing Flows and Contrastive Data

TL;DR

This paper introduces a novel outlier detection method using a contrastive normalizing flow approach that leverages an auxiliary dataset and self-supervised features to improve detection accuracy over traditional likelihood-based methods.

Contribution

It proposes a new outlier detection technique that learns the positive difference between in-distribution and contrastive feature densities using normalizing flows.

Findings

Outperforms likelihood-based outlier scores on benchmark datasets.

Effective use of contrastive data improves detection accuracy.

Method surpasses state-of-the-art anomaly detection techniques.

Abstract

Detecting test data deviating from training data is a central problem for safe and robust machine learning. Likelihoods learned by a generative model, e.g., a normalizing flow via standard log-likelihood training, perform poorly as an outlier score. We propose to use an unlabelled auxiliary dataset and a probabilistic outlier score for outlier detection. We use a self-supervised feature extractor trained on the auxiliary dataset and train a normalizing flow on the extracted features by maximizing the likelihood on in-distribution data and minimizing the likelihood on the contrastive dataset. We show that this is equivalent to learning the normalized positive difference between the in-distribution and the contrastive feature density. We conduct experiments on benchmark datasets and compare to the likelihood, the likelihood ratio and state-of-the-art anomaly detection methods.