On Label Shift in Domain Adaptation via Wasserstein Distance

TL;DR

This paper investigates label shift in domain adaptation using Wasserstein distance, proposing a new method LDROT to reduce data and label shifts, supported by theoretical analysis and extensive experiments across various settings.

Contribution

It introduces a novel optimal transport-based framework for understanding and mitigating label shift in diverse domain adaptation scenarios.

Findings

LDROT effectively reduces label and data shifts.

Theoretical analysis clarifies properties of domain adaptation under label shift.

Experimental results outperform existing baselines.

Abstract

We study the label shift problem between the source and target domains in general domain adaptation (DA) settings. We consider transformations transporting the target to source domains, which enable us to align the source and target examples. Through those transformations, we define the label shift between two domains via optimal transport and develop theory to investigate the properties of DA under various DA settings (e.g., closed-set, partial-set, open-set, and universal settings). Inspired from the developed theory, we propose Label and Data Shift Reduction via Optimal Transport (LDROT) which can mitigate the data and label shifts simultaneously. Finally, we conduct comprehensive experiments to verify our theoretical findings and compare LDROT with state-of-the-art baselines.