Graph correlated discriminant embedding for multi-source domain adaptation

Wai Keung Wong, Yuwu Lu, Zhihui Lai, Xuelong Li

Abstract: Highlights•We proposed a novel method graph correlated discriminant embedding (GCDE) for multi-source domain adaptation.•GCDE can encode the within- and between- class information of each domain data.•GCDE preserves the local and global structure information of the data.•GCDE extracts the maximization correlative features from different domains.•We also extend GCDE to a kernel case and propose kernel GCDE (KGCDE) to obtain a nonlinear representation of GCDE.