A new lattice polynomial perceptron (LPP) model is proposed which has better numerical properties than those of the direct form polynomial perceptron (PP) and of the multilayer perceptron (MLP). It shows faster and less input signal-dependent convergence behaviour and is very suitable for VLSI implementations. Experimental results are given.
Published in:
Electronics Letters
(Volume:28
,
Issue:
19
)
Date of Publication: 10 Sept. 1992