University of Cagliari
---
Abstract: Following Fong et al. (2023), Bayesian uncertainty is viewed as arising from unobserved data: with full population information, the inferential target would be known exactly. This talk focuses on the construction of prior-free posterior distributions based on one-step-ahead predictive distribution functions, which are often easier to motivate and interpret than prior distributions themselves. Recent interest has been with Hill's prediction model through what has become known as conformal prediction. In this model, the next observation is assumed to fall with equal probability into any of the intervals determined by the observed data. The resulting prediction mechanism generates complete data sets which can be used to provide posterior inference on any statistic of interest.
Il seminario potrà essere seguito anche su Microsoft Teams al link: