Monthly
288 pp. per issue
6 x 9, illustrated
ISSN
0899-7667
E-ISSN
1530-888X
2014 Impact factor:
2.21

Neural Computation

Spring 1990, Vol. 2, No. 1, Pages 116-126
(doi: 10.1162/neco.1990.2.1.116)
© 1990 Massachusetts Institute of Technology
Gram–Schmidt Neural Nets
Article PDF (423.63 KB)
Abstract

A new type of feedforward multilayer neural net is proposed that exhibits fast convergence properties. It is defined by inserting a fast adaptive Gram–Schmidt preprocessor at each layer, followed by a conventional linear combiner-sigmoid part which is adapted by a fast version of the backpropagation rule. The resulting network structure is the multilayer generalization of the gradient adaptive lattice filter and the Gram–Schmidt adaptive array.