Rethinking Approximate Gaussian Inference in Classification
Bálint Mucsányi, Nathaël Da Costa, Philipp Hennig · 2025
Read Rethinking Approximate Gaussian Inference in Classification by Bálint Mucsányi, Nathaël Da Costa, Philipp Hennig (2025) 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-2502-03366.txt.
Description
Balint Mucsanyi, Nathanael Da Costa, and Philipp Hennig rethought in 2025 how to gauge uncertainty in classification with approximate Gaussian inference. The usual softmax output shows only one kind of uncertainty, from noisy data, and misses doubt about the model itself. Gaussian methods can add that doubt, but their math needs costly sampling.
The authors propose swapping softmax for a normCDF or sigmoid step, which lets the needed values be worked out without sampling. They also match the result to a Dirichlet form for a clean final answer. Tested with several inference methods on image datasets large and small, the approach improved uncertainty estimates while removing sampling cost.
Catalog record
| CREATOR | Bálint Mucsányi, Nathaël Da Costa, Philipp Hennig |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 122 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-03366