Conference Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 51 - BFF: Innovation in Statistical Foundations
Type: Topic Contributed
Date/Time: Sunday, August 7, 2022 : 4:00 PM to 5:50 PM
Sponsor: Section on Bayesian Statistical Science
Abstract #322401
Title: Conformal Prediction with Knowledge Transfer
Author(s): Linjun Zhang*
Companies: Rutgers University
Keywords: transfer learning; conformal prediction

This talk studies conformal prediction in the setting of transfer learning, where the data in the target domain is limited but some auxiliary datasets from related source tasks are available. We propose a general framework to form prediction intervals by pooling the related data, and demonstrate the desired coverage as well as sample efficiency by applying this framework to both the standard split conformal and conformalized quantile regression (CQR) methods. Moreover, a novel meta-combination method for predictive intervals is proposed when there are communication constraints on the source tasks. We demonstrate the superiority of this approach through thorough experiments and real data applications in health care. This is joint work with Xiaotian Hou, Peng Wang and Minge Xie.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2022 program