Robust Sparse Canonical Correlation Analysis: New Formulation and Application to Fault Detection

Abstract

Canonical correlation analysis (CCA) has received wide attention in multiview representation. To improve its reliability, this letter studies a new robust sparse CCA formulation called RSCCA. Technically, a sparse matrix is introduced to characterize the sample-specific corruptions with a column-sparsity regularization term. Furthermore, an efficient alternating minimization optimization algorithm is developed. Finally, application examples on fault detection verify the superiority of the proposed RSCCA over existing CCA and sparse CCA. The results suggest that the proposed method is promising for multiview representation with robust performance.