Anthony L. Caterini

Mar 27, 2024 | arxiv.org | Brendan Leigh |Brendan Ross |Jesse C. Cresswell |Anthony L. Caterini

[Submitted on 27 Mar 2024] Title:A Geometric Explanation of the Likelihood OOD Detection Paradox View a PDF of the paper titled A Geometric Explanation of the Likelihood OOD Detection Paradox, by Hamidreza Kamkari and 5 other authors View PDF HTML (experimental) Abstract:Likelihood-based deep generative models (DGMs) commonly exhibit a puzzling behaviour: when trained on a relatively complex dataset, they assign higher likelihood values to out-of-distribution (OOD) data from simpler sources.