Title :
An upper bound on the node complexity of depth-2 multilayer perceptrons
Author_Institution :
Dept. of Inf. & Commun. Technol., Tokai Univ., Shizuoka, Japan
Abstract :
This paper shows that an upper bound on the node complexity of depth-2 perceptrons is 2N-2+2 for N-dimensional binary valued inputs. This result is obtained by structuring an N-dimensional hypercube. It is shown that the nodes are divided into the totally ordered sets of linearly independent nodes. For each set there is a hyperplane corresponding to one of the hidden units of a depth-2 perceptron which recognizes input patterns. The result of this paper is given by estimating the number of the sets
Keywords :
hypercube networks; multilayer perceptrons; neural net architecture; artificial neural networks; depth-2 multilayer perceptrons; hyperplane; linearly independent nodes; mulitdimensional binary valued inputs; multidimensional hypercube; node complexity; totally ordered sets; upper bound; Artificial neural networks; Circuits; Communications technology; Computer networks; Hypercubes; Logic; Multilayer perceptrons; Nonhomogeneous media; Pattern recognition; Upper bound;
Conference_Titel :
Neural Networks,1997., International Conference on
Conference_Location :
Houston, TX
Print_ISBN :
0-7803-4122-8
DOI :
10.1109/ICNN.1997.616192
