Conditional Distribution Estimation

The field estimates how an outcome varies with observed covariates, not just its average, so predictions can represent uncertainty, skew, and multiple plausible outcomes. Approaches include quantile regression and tree ensembles, mixture and Gaussian-process models, and neural methods such as normalizing flows, energy models, and conditional diffusion. Researchers also study how to estimate these distributions efficiently for high-dimensional, mixed-type, censored, or sequential data, and how to evaluate them with proper scores and calibration. A central tension is balancing flexible modeling of complex distributions with reliable uncertainty estimates, computational scalability, and finite-sample guarantees.