Falsehoods that ML researchers believe about OOD detection
It addresses a foundational problem in machine learning for researchers and practitioners by clarifying misconceptions and providing a unified perspective on OOD detection methods.
The paper tackles the failure of density-based out-of-distribution (OOD) detection in deep learning by listing falsehoods believed by researchers, proposing a framework to unify likelihood-ratio methods, and arguing that likelihood ratio is a principled approach rather than a fix.
An intuitive way to detect out-of-distribution (OOD) data is via the density function of a fitted probabilistic generative model: points with low density may be classed as OOD. But this approach has been found to fail, in deep learning settings. In this paper, we list some falsehoods that machine learning researchers believe about density-based OOD detection. Many recent works have proposed likelihood-ratio-based methods to `fix' the problem. We propose a framework, the OOD proxy framework, to unify these methods, and we argue that likelihood ratio is a principled method for OOD detection and not a mere `fix'. Finally, we discuss the relationship between domain discrimination and semantics.