A Fast deflation Method for Sparse Principal Component Analysis via Subspace Projections
share
Deflation method is an iterative technique that searches the sparse loadings one by one. However, the dimensionality of the search space is usually fixed in each step, as the same as that of the original data, which brings heavy cumulative computational cost in high-dimensional statistics. To address this problem, we propose a fast deflation method via subspace projections. By using Household QR factorization, a series of subspace projections are constructed to restrict the computation of each loading in a low dimensional subspace orthogonal to the previous sparse loadings. Experimental results demonstrate that the proposed method acheives the state-of-the-art performance on benchmark data sets and gets a significant improvement on computational speed.
READ FULL TEXT
