Importance Weight Estimation and Generalization in Domain Adaptation under Label Shift

TL;DR

This paper investigates how to estimate importance weights for domain adaptation under label shift, providing methods with confidence bounds and analyzing their impact on generalization in unlabeled target domains.

Contribution

It introduces new importance weight estimators with confidence bounds and analyzes their role in improving generalization under label shift.

Findings

Proposed importance weight estimators with confidence bounds.

Derived generalization bounds for unlabeled target domains.

Demonstrated effectiveness of estimators in domain adaptation scenarios.

Abstract

We study generalization under labeled shift for categorical and general normed label spaces. We propose a series of methods to estimate the importance weights from labeled source to unlabeled target domain and provide confidence bounds for these estimators. We deploy these estimators and provide generalization bounds in the unlabeled target domain.