Title: Calibrated Predictive Distributions via Diagnostics for Conditional Coverage

Abstract: Uncertainty quantification is crucial for assessing the predictive ability of AI algorithms. A large body of work (including normalizing flows and Bayesian neural networks) has been devoted to describing the entire predictive distribution (PD) of a target variable Y given input features $\mathbf{X}$. However, off-the-shelf PDs are usually far from being conditionally calibrated; i.e., the probability of occurrence of an event given input $\mathbf{X}$ can be significantly different from the predicted probability. Most current research on predictive inference (such as conformal prediction) concerns constructing calibrated prediction sets only. It is often believed that the problem of obtaining and assessing entire conditionally calibrated PDs is too challenging. In this work, we show that recalibration, as well as diagnostics of entire PDs, are indeed attainable goals in practice. Our proposed method relies on the idea of regressing probability integral transform (PIT) scores against $\mathbf{X}$. This regression gives full diagnostics of conditional coverage across the entire feature space and can be used to recalibrate misspecified PDs. We benchmark our corrected prediction bands against oracle bands and state-of-the-art predictive inference algorithms for synthetic data, including settings with a distributional shift. Finally, we produce calibrated PDs for two applications: (i) probabilistic nowcasting based on sequences of satellite images, and (ii) estimation of galaxy distances based on imaging data (photometric redshifts).