Title: Assessing the Uncertainty Quantification Capabilities of Tabular Foundation Models

Original: arXiv:2606.01427v1 Announce Type: cross Abstract: Foundation models (FMs) have achieved substantial success in generalizing across tasks without problemspecific training or fine-tuning. However, many critical applications in mechanics and computational science require not only accurate predictions but also reliable uncertainty quantification (UQ). Herein we investigate the UQ capabilities of tabular FMs in regression tasks through a comprehensive empirical study comparing Tabular Prior-Data Fitted Networks (TabPFN) against Gaussian processes (GPs). We systematically evaluate these two methods across a host of regression problems with varying complexity, dataset sizes, and input dimensionalities. We use a default setting to build all the GPs and for a fair comparison against TabPFN v2.5. Our findings highlight an important trade-off between explicit and learned priors: while TabPFN achieves highly competitive performance for complex, high-dimensional problems with sufficient data, GPs often provide superior predictive accuracy and UQ in data-scarce settings. Moreover, when the chosen kernel constitutes a good prior for the underlying function, GP performance can substantially exceed that of TabPFN. Our results can be reproduced from https://github.com/kianswarehouse/GPvsPFN.

Rewrite: Title: Evaluating Uncertainty Quantification in Tabular Foundation Models

Abstract: Foundation models (FMs) have demonstrated remarkable proficiency in generalizing across various tasks, eliminating the need for task-specific training or fine-tuning. Nevertheless, numerous vital domains within mechanics and computational science demand not just precise predictions, but also trustworthy uncertainty quantification (UQ). This study explores the UQ potential of tabular FMs in regression scenarios via an extensive empirical analysis that juxtaposes Tabular Prior-Data Fitted Networks (TabPFN) with Gaussian processes (GPs). The evaluation encompasses a wide array of regression challenges characterized by differing levels of complexity, dataset magnitudes, and input dimensions. For a balanced comparison with TabPFN v2.5, all GPs were constructed using default configurations. The results underscore a significant compromise between explicit and learned priors: TabPFN delivers competitive outcomes for intricate, high-dimensional issues when adequate data is available; conversely, GPs tend to offer better predictive precision and UQ in contexts with limited data. Furthermore, if the selected kernel serves as an appropriate prior for the underlying function, GP performance can markedly surpass that of TabPFN. Reproducibility of these findings is ensured via https://github.com/kianswarehouse/GPvsPFN.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC