Page 449 - Jolliffe I. Principal Component Analysis
P. 449

Appendix A. Computation of Principal Components
                              414
                                Neural network algorithms are feasible for larger data sets than batch
                              methods because they are better able to take advantage of developments
                              in computer architecture. DK96, Chapter 8, discuss the potential for ex-
                              ploiting parallel VSLI (very large scale integration) systems, where the
                              most appropriate algorithms may be different from those for non-parallel
                              systems (DK96, Section 3.5.5). They discuss both digital and analogue
                              implementations and their pros and cons (DK96, Section 8.3). Classical
                              eigenvector-based algorithms are not easily parallelizable, whereas neural
                              network algorithms are (DK96 pp. 205–207).
   444   445   446   447   448   449   450   451   452   453   454