Domain Adaptation with Incomplete Target Domains

Zhenpeng Li,Jianan Jiang,Yuhong Guo,Tiantian Tang,Chengxiang Zhuo,Jieping Ye
2023-06-13
Abstract:Domain adaptation, as a task of reducing the annotation cost in a target domain by exploiting the existing labeled data in an auxiliary source domain, has received a lot of attention in the research community. However, the standard domain adaptation has assumed perfectly observed data in both domains, while in real world applications the existence of missing data can be prevalent. In this paper, we tackle a more challenging domain adaptation scenario where one has an incomplete target domain with partially observed data. We propose an Incomplete Data Imputation based Adversarial Network (IDIAN) model to address this new domain adaptation challenge. In the proposed model, we design a data imputation module to fill the missing feature values based on the partial observations in the target domain, while aligning the two domains via deep adversarial adaption. We conduct experiments on both cross-domain benchmark tasks and a real world adaptation task with imperfect target domains. The experimental results demonstrate the effectiveness of the proposed method.
Machine Learning,Artificial Intelligence
What problem does this paper attempt to address?
The paper attempts to address the problem of domain adaptation in the presence of incomplete target domain data. Specifically, traditional domain adaptation methods assume that data from both the source and target domains are fully observable. However, in practical applications, data in the target domain often contains missing values. In such cases, directly applying standard domain adaptation methods may fail to produce satisfactory results due to the neglect of data incompleteness. Therefore, the paper proposes a novel adversarial domain adaptation model—Incomplete Data Imputation Adversarial Network (IDIAN), to tackle the challenges posed by incomplete target domain data. The main objective of IDIAN is to train an effective classifier in the target domain by fully leveraging the fully observable and labeled data in the source domain. The model is designed to handle both homogeneous and heterogeneous cross-domain feature spaces and operates in a semi-supervised setting. By introducing a data generator to fill in the missing values in the target domain and using adversarial methods to align the feature distributions of the source and target domains, the model achieves effective knowledge transfer and target domain classifier training. Experimental results demonstrate that, compared to existing adversarial domain adaptation methods, the model shows effectiveness in both simulated incomplete target domains and real-world application scenarios.