Out-of-Distribution Detection with Class Ratio Estimation

TL;DR

This paper introduces a new framework for out-of-distribution detection that unifies existing density ratio methods using energy-based models and class ratio estimation, achieving competitive results without complex generative models.

Contribution

It proposes a novel probabilistic framework unifying density ratio methods with energy-based models and introduces class ratio estimation for improved OOD detection.

Findings

Competitive OOD detection results on image datasets

Framework unifies and explains existing density ratio approaches

Avoids complex deep generative models for OOD detection

Abstract

Density-based Out-of-distribution (OOD) detection has recently been shown unreliable for the task of detecting OOD images. Various density ratio based approaches achieve good empirical performance, however methods typically lack a principled probabilistic modelling explanation. In this work, we propose to unify density ratio based methods under a novel framework that builds energy-based models and employs differing base distributions. Under our framework, the density ratio can be viewed as the unnormalized density of an implicit semantic distribution. Further, we propose to directly estimate the density ratio of a data sample through class ratio estimation. We report competitive results on OOD image problems in comparison with recent work that alternatively requires training of deep generative models for the task. Our approach enables a simple and yet effective path towards solving the OOD detection problem.