Synthetic-powered predictive inference

Edgar Dobriban (University of Pennsylvania) · Yaniv Romano (Technion---Israel Institute of Technology) · Meshi Bashari (Technion) · Roy Maor Lotan (Technion - Israel Institute of Technology) · Yonghoon Lee (The Wharton School, University of Pennsylvania)

calibration processconformal prediction data augmentation diffusion modelempirical quantile mappingfinite-sample coveragefinite-sample guaranteegenerative modelnonconformity scoresprediction setspredictive efficiencypredictive inferencesample efficiencyscore transportersynthetic data

Conformal prediction is a framework for predictive inference with a distribution-free, finite-sample guarantee. However, it tends to provide uninformative prediction sets when calibration data are scarce. This paper introduces Synthetic-powered predictive inference (SPI), a novel framework that incorporates synthetic data