Data-Driven Approach to Multiple-Source Domain Adaptation

Abstract

A key problem in domain adaptation is determining what to transfer across different domains. We propose a data-driven method to represent these changes across multiple source domains and perform unsupervised domain adaptation. We assume that the joint distributions follow a specific generating process and have a small number of identifiable changing parameters, and develop a data-driven method to identify the changing parameters by learning low-dimensional representations of the changing class-conditional distributions across multiple source domains. The learned low-dimensional representations enable us to reconstruct the target-domain joint distribution from unlabeled target-domain data, and further enable predicting the labels in the target domain. We demonstrate the efficacy of this method by conducting experiments on synthetic and real datasets.

Figure 1:

Generating process Y → X across domains with domain index variable D = 1, …, M

Figure 2:

Accuracies of the baselines and the proposed method for the task of classifying between digits 4 and 9, for handwritten digit recognition

Figure 3:

Accuracies of the baselines and the proposed method for the task of classifying between digits 1 and 7, for handwritten digit recognition

Figure 4:

Accuracies of the baselines and the proposed method for the task of classifying between two different lung fissures in the real dataset