Self-Supervised Conformal Prediction with Equivariant Bootstrapping for Image Uncertainty Quantification

2026-05-18

AI Summary

This work addresses the challenge of reliably quantifying reconstruction uncertainty in ill-posed imaging inverse problems, where existing methods often introduce bias in the absence of ground-truth labels. To overcome this limitation, we propose an unsupervised uncertainty quantification framework that requires no calibrated ground-truth data. Our approach uniquely integrates equivariant bootstrapping with self-supervised conformal prediction to produce high-confidence prediction intervals, demonstrated effectively in the context of weak gravitational lensing mass mapping. By mitigating biases induced by distributional shifts, the method is particularly well-suited for scientific imaging scenarios where ground-truth labels are scarce or model-dependent. The proposed framework significantly enhances both the reliability and interpretability of reconstructed images.

Abstract

Inverse problems are ubiquitous in modern scientific studies and involve recovering an underlying signal from noisy observations often transformed by a measurement operator. These problems are frequently ill-posed, particularly in imaging, leading to multiple plausible solutions and considerable uncertainty in reconstructed images. In fields like the physical and biological sciences, accurate uncertainty quantification (UQ) is critical for trustworthy scientific analyses and confident diagnoses. Current UQ methods for imaging often fall short; they can be inaccurate, or require unavailable or difficult-to-acquire ground truth data for calibration, which can introduce hidden biases due to distribution shifts between calibration and observed data. We introduce a UQ approach that leverages equivariant bootstrapping to generate heuristic coverages by exploiting data symmetries. We then refine these coverages through a conformal prediction calibration step, while crucially employing a self-supervised approach to avoid the need for ground truth calibration data. We demonstrate this method with weak lensing mass-mapping, where we aim to reconstruct the convergence field from shear measurements of distant galaxies weakly-lensed by gravitational fields. Mass-mapping in particular benefits from the self-supervised approach, as simulating calibration data is expensive and relies on specific cosmological models that could introduce biases in downstream cosmological inference tasks.

Problem

• Uncertainty Quantification
• Inverse Problems
• Imaging
• Conformal Prediction
• Self-Supervised Learning

Innovation

• Self-Supervised Learning
• Conformal Prediction
• Equivariant Bootstrapping
• Uncertainty Quantification
• Inverse Problems