Domain regularized transfer component analysis

Anti-drift is an emergent and challenging issue in data-related subjects. In this paper, we propose to address the time-varying drift (e.g. electronic nose drift). By viewing drift to be with different probability distribution from the regular data, a robust subspace projection approach with PCA synthesis is proposed for anti-drift. The main idea behind is that given two clusters of data points with different probability distribution caused by drift, we tend to find a latent projection P (i.e. a group of basis), such that the newly projected subspace of the two clusters is with similar distribution (i.e. anti-drift). The merits of the proposed domain regularized component analysis (DRCA) method are threefold: 1) the proposed subspace projection mechanism is unsupervised, without using any label information of data in anti-drift; 2) a simple but effective concept of domain distance is proposed to represent the mean distribution discrepancy metric; 3) the proposed anti-drift method can be easily solved by Eigen decomposition, and anti-drift is manifested with a well solved projection matrix in real time application. Experiment on a benchmark drift dataset demonstrates the effectiveness of the proposed DRCA method.