Adaptive kernel subspace method for speeding up feature extraction

KPCA can extract nonlinear features of data set. However, its efficiency is in inverse proportion to the size of the training sample set. In this paper, we proposed an adaptive kernel subspace method to extract features efficiently. The method is methodologically consistent with KPCA, and can improve the efficiency by adaptively selecting the spanning vectors of the kernel principal components, meanwhile, not affect the accuracy much. Experiments on two-dimensional data, MNIST dataset and USPS dataset show that the feature extraction method is more efficient than that associated with KPCA and reference methods.