Fast nyström for low rank matrix approximation

Abstract

Low-rank matrix approximation is a crucial technique for data analysis and scientific computing, and the Nyström method is one of the efficient sampling-based low-rank approximation schemes for handling large kernel matrices. The approximation accuracy of Nyström approach highly depends on the number of columns of the subset used, and it consumes much time on large data sets. This paper presents an accurate and fast Nyström approach to reducing the computational burdens when handling large kernel matrices, and experimental results show its competitive performance in both accuracy and efficiency. © Springer-Verlag Berlin Heidelberg 2012.