Testing Exchangeability for Transfer Decision

Shuang Zhou*, Evgueni Smirnov, Gijsbertus Schoenmakers, Kurt Driessens, Ralf Peeters

*Corresponding author for this work

Research output: Contribution to journalArticleAcademicpeer-review

Abstract

This paper introduces a non-parametric test to decide whether to transfer data from a source domain to a target domain to improve the generalization performance of predictive models on the target domain. The test is based on the conformal prediction framework: it statistically tests whether the target and source data are generated from the same distribution under the exchangeability assumption. The experiments show that the test is capable of outperforming existing methods when it decides on instance transfer. 

Original languageEnglish
Pages (from-to)64-71
Number of pages8
JournalPattern Recognition Letters
Volume88
DOIs
Publication statusPublished - 1 Mar 2017

Keywords

  • Conformity prediction framework
  • Exchangeability test
  • Instance-transfer learning
  • SELECTION

Cite this