Timezone: »

Partial disentanglement for domain adaptation
Lingjing Kong · Shaoan Xie · Weiran Yao · Yujia Zheng · Guangyi Chen · Petar Stojanov · Victor Akinwande · Kun Zhang

Tue Jul 19 08:55 AM -- 09:00 AM (PDT) @ Room 318 - 320

Unsupervised domain adaptation is critical to many real-world applications where label information is unavailable in the target domain. In general, without further assumptions, the joint distribution of the features and the label is not identifiable in the target domain. To address this issue, we rely on a property of minimal changes of causal mechanisms across domains to minimize unnecessary influences of domain shift. To encode this property, we first formulate the data generating process using a latent variable model with two partitioned latent subspaces: invariant components whose distributions stay the same across domains, and sparse changing components that vary across domains. We further constrain the domain shift to have a restrictive influence on the changing components. Under mild conditions, we show that the latent variables are partially identifiable, from which it follows that the joint distribution of data and labels in the target domain is also identifiable. Given the theoretical insights, we propose a practical domain adaptation framework, called iMSDA. Extensive experimental results reveal that iMSDA outperforms state-of-the-art domain adaptation algorithms on benchmark datasets, demonstrating the effectiveness of our framework.

Author Information

Lingjing Kong (Carnegie Mellon University)
Shaoan Xie (Carnegie Mellon University)
Weiran Yao (Carnegie Mellon University)
Yujia Zheng (Carnegie Mellon University)
Guangyi Chen (MBZUAI)
Petar Stojanov (Broad Institute of MIT and Harvard)
Victor Akinwande (Carnegie Mellon University)
Kun Zhang (Carnegie Mellon University)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors