Abstract:Conventional domain adaptation methods do not work well when a large gap exists between the source and the target domain. Gradual domain adaptation is one of the approaches to address the problem by leveraging the intermediate domain, which gradually shifts from the source to the target domain. The previous work assumed that the number of the intermediate domains is large and the distance of the adjacent domains is small; hence, the gradual domain adaptation algorithm by self-training with unlabeled datasets was applicable. In practice, however, gradual self-training will fail because the number of the intermediate domains is limited, and the distance of the adjacent domains is large. We propose using normalizing flows to mitigate this problem while maintaining the framework of unsupervised domain adaptation. We generate pseudo intermediate domains from normalizing flows and then use them for gradual domain adaptation. We evaluate our method by experiments with real-world datasets and confirm that our proposed method mitigates the above explained problem and improves the classification performance.
Abstract:In domain adaptation, when there is a large distance between the source and target domains, the prediction performance will degrade. Gradual domain adaptation is one of the solutions to such an issue, assuming that we have access to intermediate domains, which shift gradually from the source to target domains. In previous works, it was assumed that the number of samples in the intermediate domains is sufficiently large; hence, self-training was possible without the need for labeled data. If access to an intermediate domain is restricted, self-training will fail. Practically, the cost of samples in intermediate domains will vary, and it is natural to consider that the closer an intermediate domain is to the target domain, the higher the cost of obtaining samples from the intermediate domain is. To solve the trade-off between cost and accuracy, we propose a framework that combines multifidelity and active domain adaptation. The effectiveness of the proposed method is evaluated by experiments with both artificial and real-world datasets. Codes are available at https://github.com/ssgw320/gdamf.