Contrastive Training for Improved Out-of-Distribution Detection

TL;DR

This paper introduces a contrastive training approach that enhances out-of-distribution detection in machine learning systems without needing explicit OOD examples, especially improving performance on challenging near OOD cases.

Contribution

It presents a novel contrastive training method for OOD detection that does not require labeled OOD data, and introduces the CLP score to measure detection difficulty.

Findings

Contrastive training significantly improves OOD detection performance.

The method especially benefits near OOD class detection.

The CLP score effectively quantifies OOD detection difficulty.

Abstract

Reliable detection of out-of-distribution (OOD) inputs is increasingly understood to be a precondition for deployment of machine learning systems. This paper proposes and investigates the use of contrastive training to boost OOD detection performance. Unlike leading methods for OOD detection, our approach does not require access to examples labeled explicitly as OOD, which can be difficult to collect in practice. We show in extensive experiments that contrastive training significantly helps OOD detection performance on a number of common benchmarks. By introducing and employing the Confusion Log Probability (CLP) score, which quantifies the difficulty of the OOD detection task by capturing the similarity of inlier and outlier datasets, we show that our method especially improves performance in the `near OOD' classes -- a particularly challenging setting for previous methods.