SPACR: Single-Pass Adaptive Training of Uncertainty-Aware Conformal Regressors

2026-06-09

AI Summary

This work addresses the misalignment between model training objectives and the goal of producing efficient (i.e., narrow) prediction intervals in traditional conformal prediction (CP), which is typically applied as a post-hoc procedure. The authors propose SPACR, a novel method that, for the first time, jointly optimizes both validity and efficiency of prediction intervals within a single end-to-end training process via a differentiable loss function—eliminating the need for data splitting or pre-specifying multiple confidence levels. SPACR enables a single model to output valid prediction intervals simultaneously across multiple confidence levels, avoiding the repeated training required by approaches like DOICR and thereby substantially improving computational efficiency and interval sharpness. Experiments demonstrate that SPACR consistently yields narrower yet well-calibrated intervals across diverse datasets, achieving a superior trade-off among coverage accuracy, efficiency, and computational cost.

Abstract

Conformal Prediction (CP) provides robust uncertainty guarantees for predictive models, but is typically applied post hoc, which misaligns model training with the conformal goal of producing efficient (i.e, narrow) intervals. We propose SPACR (Single-Pass Adaptive Conformal Regressor), a novel method for directly training uncertainty-aware regressors within a differentiable loss. SPACR jointly optimizes efficiency and validity without batch-splitting or a predefined confidence levels during training. As a result, a single SPACR model yields valid prediction intervals at multiple confidence levels during inference, avoiding the costly retraining required by methods like DOICR. Experiments on diverse datasets show that SPACR consistently gives tighter intervals and better coverage-efficiency trade-offs compared to standard CP and DOICR, while significantly reducing computational costs.

Problem

• Conformal Prediction
• Uncertainty-aware Regression
• Prediction Intervals
• Efficiency-Validity Trade-off
• Post-hoc Calibration

Innovation

• Conformal Prediction
• Uncertainty-Aware Regression
• Single-Pass Training
• Differentiable Loss
• Adaptive Confidence Levels