Toward Reliable Domain Adaptation under Continuous Spurious Shift

Yi Wang, Zihao Xu, Hao He, Chengzhi Mao, Qi Xu, Zhigang Hua, Hao Wang

Keywords: Probabilistic Models; Continuous Shift

Abstract: Recent advances in domain adaptation have shown promise in transferring knowledge across domains characterized by a continuous value or vector, such as varying patient ages, where "age'' serves as a continuous index. However, these approaches often fail when spurious features shift continuously along with the domain index. This paper introduces the first method designed to withstand the continuous shifting of spurious features during domain adaptation. Our method enhances domain adaptation performance by aligning causally transportable encodings across continuously indexed domains. Theoretical analysis demonstrates that our approach more effectively ensures causal transportability across different domains. Empirical results, from both semi-synthetic and real-world medical datasets, indicate that our method outperforms state-of-the-art domain adaptation methods.

Primary Area: probabilistic methods (Bayesian methods, variational inference, sampling, UQ, etc.)

Submission Number: 23328