Rethinking Vacuity for OOD Detection in Evidential Deep Learning
Claire McNamara · 2026
Read Rethinking Vacuity for OOD Detection in Evidential Deep Learning by Claire McNamara (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-06382.txt.
Description
Claire McNamara shows that a popular uncertainty measure for evidential deep learning quietly breaks when class counts differ. Vacuity, also called uncertainty mass, scores out of distribution detection by dividing the number of classes by the total belief strength of a prediction. Because the class count sits inside the measure, comparing test sets whose in and out of distribution sides use different counts can distort the verdict. It can do so without any change in the model's actual predictions, as McNamara demonstrates empirically and theoretically.
Swapping class cardinality by just one moved detection scores dramatically in her tests on evidential models fine tuned from a large language model. The shift is an evaluation artifact, inflating results through arithmetic rather than insight. McNamara also questions how in and out of distribution are defined for multiple choice question answering with causal language models. The contribution is a caution the subfield needs.
Catalog record
| CREATOR | Claire McNamara |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 42 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.06382