Pp. Kanjilal et al., REDUCED-SIZE NEURAL NETWORKS THROUGH SINGULAR-VALUE DECOMPOSITION ANDSUBSET-SELECTION, Electronics Letters, 29(17), 1993, pp. 1516-1518
Singular value decomposition (SVD) can lead to compaction of informati
on, and along with QR with column pivoting (QRcp) factorisation can be
used for subset selection. The Letter proposes an application of SVD
and subset selection for optimising the size of feedforward neural net
works; the Mackey Glass (MG) series is used as an example.