Domain Adaptation Based Multi-Source Data Fusion for Pattern Classification

Abstract

Multi-source Unsupervised Domain Adaptation (MUDA) is an important and challenging research topic for target classification with the assistance of labeled data in source domains. When we have several labeled source domains, it is hard to map all source domains and target domain into a common feature space for well classifying the targets. A new Progressive Multi-Source Domain Adaptation Network (PMSDAN) is proposed to further improve the classification performance. PMSDAN mainly consists of two steps for distribution alignment. Firstly, the multiple source domains are integrated as one auxiliary domain to match the distribution with the target domain. In order to mine as much as possible assistance knowledge from each source domain, the distribution of the target domain will be separately aligned with that of each source domain. Finally, a weighted fusion method is employed to combine the multiple classification results for making the final classification decision. In the optimization of domain adaption, Weighted Hybrid Maximum Mean Discrepancy (WHMMD) is proposed, and it considers both the inter-class and intra-class Discrepancy. The experiments show that the proposed method can obtain remarkable results for MUDA on public benchmark dataset compared to some state-of-the-art methods.