Conformal Prediction and Conditional Coverage for Tabular Foundation Models

2026-09-28

AI Summary

This study addresses the under-coverage and over-coverage issues in prediction intervals generated by tabular foundation models by proposing C-USIM. Built upon highest predictive density split conformal prediction, this method produces multimodal prediction intervals without requiring additional model training. Its core innovation lies in introducing a lightweight adaptation mechanism to handle multimodal distributions, providing finite-sample marginal coverage guarantees while theoretically quantifying the conditional-marginal coverage gap. Experimental results demonstrate that C-USIM significantly improves marginal coverage accuracy on models such as TabPFN and effectively reduces both conditional and group coverage errors.

Abstract

Tabular foundation models (TFMs) provide predictive distributions for regression, but their prediction regions can exhibit undercoverage or overcoverage even when point predictions are accurate. We introduce C-USIM (Conditionally-Uniformized Score Integration Method), a lightweight application of highest predictive density split conformal prediction that accommodates multimodal predictions. Given calibration and test outputs, it requires no additional training or model inference. It provides finite-sample marginal validity under our assumptions. We bound conditional-marginal coverage gaps using distribution-estimation error and score discreteness, and examine coverage heterogeneity through percentile rank-score plots. Experiments with TabPFN and TabICL show improved marginal coverage accuracy and lower average conditional and group coverage errors. Under a fixed data budget, allocating more observations to calibration can reduce marginal coverage error despite less accurate point predictions.

Problem

• Tabular Foundation Models
• Conformal Prediction
• Conditional Coverage
• Regression
• Prediction Intervals

Innovation

• Conformal Prediction
• Tabular Foundation Models
• Conditional Coverage
• C-USIM
• Highest Predictive Density