Sparse discriminant PCA based on contrastive learning and class-specificity distribution

Qian Zhou, Quanxue Gao, Qianqian Wang, Ming Yang, Xinbo Gao

Abstract: Highlights•Contrastive learning enhances PCA by incorporating both individual sample reconstruction and discriminative inter-sample information, thus improving PCA’s discriminative capabilities.•The class-specific distribution of the data is explored by minimizing the squared ℓ1,2-norm of the projected data, which facilitates further extraction of discriminant information•The PCA algorithm is made more interpretable by flexible squared ℓ1,2-norm sparsity constraints.