Adversarial Unsupervised Domain Adaptation with Conditional and Label Shift: Infer, Align and Iterate

TL;DR

This paper introduces a novel adversarial unsupervised domain adaptation method that explicitly addresses conditional and label shifts by iteratively inferring label distributions and aligning class-conditional features, improving adaptation performance.

Contribution

It proposes a new optimization scheme for adversarial UDA that infers label marginals and aligns class-conditional distributions, handling label shifts effectively.

Findings

Effective on classification and segmentation UDA tasks

Outperforms conventional methods under label shift scenarios

Demonstrates robustness on partial UDA settings

Abstract

In this work, we propose an adversarial unsupervised domain adaptation (UDA) approach with the inherent conditional and label shifts, in which we aim to align the distributions w.r.t. both $p(x|y)$ and $p(y)$. Since the label is inaccessible in the target domain, the conventional adversarial UDA assumes $p(y)$ is invariant across domains, and relies on aligning $p(x)$ as an alternative to the $p(x|y)$ alignment. To address this, we provide a thorough theoretical and empirical analysis of the conventional adversarial UDA methods under both conditional and label shifts, and propose a novel and practical alternative optimization scheme for adversarial UDA. Specifically, we infer the marginal $p(y)$ and align $p(x|y)$ iteratively in the training, and precisely align the posterior $p(y|x)$ in testing. Our experimental results demonstrate its effectiveness on both classification and segmentation UDA, and partial UDA.