Interventional Domain Adaptation

TL;DR

This paper introduces a novel domain adaptation method that intervenes in feature learning to eliminate source-specific biases, improving the safety and transferability of discriminative features across domains.

Contribution

It proposes a feature intervention strategy using counterfactual features to prevent transfer of domain-specific biases, enhancing domain adaptation performance.

Findings

Improves transferability of discriminative features across domains.

Achieves state-of-the-art results on unsupervised domain adaptation tasks.

Enhances generalization to unseen domains.

Abstract

Domain adaptation (DA) aims to transfer discriminative features learned from source domain to target domain. Most of DA methods focus on enhancing feature transferability through domain-invariance learning. However, source-learned discriminability itself might be tailored to be biased and unsafely transferable by spurious correlations, \emph{i.e.}, part of source-specific features are correlated with category labels. We find that standard domain-invariance learning suffers from such correlations and incorrectly transfers the source-specifics. To address this issue, we intervene in the learning of feature discriminability using unlabeled target data to guide it to get rid of the domain-specific part and be safely transferable. Concretely, we generate counterfactual features that distinguish the domain-specifics from domain-sharable part through a novel feature intervention strategy. To prevent the residence of domain-specifics, the feature discriminability is trained to be invariant to the mutations in the domain-specifics of counterfactual features. Experimenting on typical \emph{one-to-one} unsupervised domain adaptation and challenging domain-agnostic adaptation tasks, the consistent performance improvements of our method over state-of-the-art approaches validate that the learned discriminative features are more safely transferable and generalize well to novel domains.