The Accelerated Power Method for Kernel Principal Component Analysis.

When faced with the large-scale data set, Kernel principal component analysis (KPCA) is infeasible because of the storage and computational problem. To overcome these disadvantages, an accelerated power method of computing kernel principal components is proposed. First, the accelerated Power iteration is introduced to compute the first eigenvalue and corresponding eigenvector. Then the deflation method is repeatedly applied to achieve other higher order eigenvectors. The space and time complexity of the proposed method is greatly reduced. Experimental results confirm the effectiveness of proposed method.