Machine Learning

This study evaluates whether frozen tabular foundation models can function as probabilistic photometric-redshift estimators for quasars …

CP4SBI is a conformal calibration framework for simulation-based inference: it recalibrates posterior credible sets from neural estimators to guarantee local Bayesian coverage. Two variants, based on regression trees and on CDFs, give finite-sample guarantees and substantially improve uncertainty quantification for estimators built with normalizing flows and diffusion models.

Generative models are increasingly used for scientific inverse problems, but their posterior or predictive uncertainty can still be …

WE introduce EPICSCORE, a model-agnostic method that enhances conformal prediction by integrating epistemic uncertainty. Compatible with any Bayesian model and maintaining distribution-free guarantees, EPICSCORE adapts prediction intervals based on data availability, achieving both finite-sample marginal and asymptotic conditional coverage.