Two Classes of Accelerated Manifold Proximal Gradient Method for Sparse PCA

Principal component analysis (PCA) is a popular technique for machine learning and signal processing. Sparse PCA is a special and difficult case of PCA and sparsity can enhance the interpretability of pricipal component. We focus on such the sparse PCA problem, which is a typical nonsmooth and nonconvex optimization problem. To the best of our knowledge, the existing methods lack of sufficient rationality in the design of algorithm or have the disvantage on speed and iteration. In this paper, we proposed two classes of accelerated manifold proximal gradient method. Numerical experiments display the superiority of our algorithms compared to the latest methods.