Uncertainty Quantification for Prior-Data Fitted Networks using Martingale Posteriors
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866914541003079680 |
|---|---|
| author | Nagler, Thomas Rügamer, David |
| author_facet | Nagler, Thomas Rügamer, David |
| contents | Prior-data fitted networks (PFNs) have emerged as promising foundation models for prediction from tabular datasets, achieving state-of-the-art performance on small to moderate data sizes without tuning. While PFNs are motivated by Bayesian ideas, they do not provide any uncertainty quantification for predictive means, quantiles, or similar quantities. We propose a principled, efficient, and tuning-free sampling procedure to construct Bayesian posteriors for such estimates based on martingale posteriors, and prove its convergence. Several simulated and real-world data examples showcase the efficiency and calibration of our method in inference applications. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_11325 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Uncertainty Quantification for Prior-Data Fitted Networks using Martingale Posteriors Nagler, Thomas Rügamer, David Methodology Artificial Intelligence Machine Learning Computation Prior-data fitted networks (PFNs) have emerged as promising foundation models for prediction from tabular datasets, achieving state-of-the-art performance on small to moderate data sizes without tuning. While PFNs are motivated by Bayesian ideas, they do not provide any uncertainty quantification for predictive means, quantiles, or similar quantities. We propose a principled, efficient, and tuning-free sampling procedure to construct Bayesian posteriors for such estimates based on martingale posteriors, and prove its convergence. Several simulated and real-world data examples showcase the efficiency and calibration of our method in inference applications. |
| title | Uncertainty Quantification for Prior-Data Fitted Networks using Martingale Posteriors |
| topic | Methodology Artificial Intelligence Machine Learning Computation |
| url | https://arxiv.org/abs/2505.11325 |