High Performance, Low Reliability: Uncertainty Benchmarking for Tabular Foundation Models
José Lucas De Melo Costa, Fabrice Popineau, Arpad Rimmel, Bich-liên Doan CentraleSupélec, Université Paris-Saclay Gif-sur-Yvette - France · 2026
Read High Performance, Low Reliability: Uncertainty Benchmarking for Tabular Foundation Models by José Lucas De Melo Costa, Fabrice Popineau, Arpad Rimmel, Bich-liên Doan CentraleSupélec, Université Paris-Saclay Gif-sur-Yvette - France (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-2605-28554.txt.
Description
José Lucas De Melo Costa and three colleagues probe uncertainty in tabular foundation models, whose prediction strength has outrun trust checks. Testing foundation models, gradient-boosted trees, and older rivals on 112 TALENT data sets, they find a trade-off: foundation models lead on raw scores but trail on case-by-case cover under conformal prediction. Made-up tests map where the gap widens. They conclude that well-calibrated uncertainty stays a major open challenge for reliable use of these models.
Catalog record
| CREATOR | José Lucas De Melo Costa, Fabrice Popineau, Arpad Rimmel, Bich-liên Doan CentraleSupélec, Université Paris-Saclay Gif-sur-Yvette - France |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 13266 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.28554