Approximations to the Fisher Information Metric of Deep Generative Models for Out-Of-Distribution Detection

arXiv:2403.01485v1 Announce Type: cross
Abstract: Likelihood-based deep generative models such as score-based diffusion models and variational autoencoders are state-of-the-art machine learning models approximating high-dimensional distributions of data such as images, text, or audio. One of many downstream tasks they can be naturally applied to is out-of-distribution (OOD) detection. However, seminal work by Nalisnick et al. which we reproduce showed that deep generative models consistently infer higher log-likelihoods for OOD data than data they were trained on, marking an open problem. …

abstract art arxiv audio autoencoders cs.cv cs.lg data deep generative models detection diffusion diffusion models distribution fisher generative generative models images information likelihood machine machine learning machine learning models state stat.ml tasks text type variational autoencoders