<ul><li>Reliably characterizing the full conditional distribution of a multivariate response variable given a set of covariates is crucial for trustworthy decision-making.</li><li>Standard recalibration methods are limited to univariate settings, and conformal prediction techniques do not provide a full probability density function for multivariate prediction regions.</li><li>A novel latent recalibration (LR) method is introduced, which assesses probabilistic calibration in the latent space of a conditional normalizing flow and provides recalibrated distribution with an explicit multivariate density function.</li><li>Extensive experiments on tabular and image datasets demonstrate that LR consistently improves latent calibration error and the negative log-likelihood of the recalibrated models.</li></ul>

Multivariate Latent Recalibration for Conditional Normalizing Flows

Discover more