On General Weighted Adaptive Sparse Principal Component Analysis

In this study, we propose a novel method termed general weighted adaptive sparse principal component analysis (GWASPCA), which employs a general weighted adaptive penalty function in place of the Lasso penalty used in sparse principal component analysis (SPCA) as presented by Zou et al. For the non-convex penalty involved, we use the local linear approximation (LLA) to dramatically reduce the computational cost. Our proposed algorithms are demonstrated superiority in terms of sparsity and variance contribution on both synthetic and real-world datasets.