Título: Issues on the Complexy of Training Weightless Neural Networks
Autores: Souto, Marcílio C. P. de; Guimarães, Katia S.; Ludermir, Teresa B.
Resumo: In this paper, it is extended the Judd’s results with respect to learning computacional complexity of weighted neural models to include the weightless neural models. It is shown, for example, that also is NP-complete any algorithm that aims to load any performable training set in any conceivable weightless neural network. It is also conjectured that a specific architecture class, pyramidal architectures (for the weightless models), may be a way to overcome the NP-completeness of learning.
Código DOI: 10.21528/CBRN1994-002
Artigo em PDF: CBRN1994-paper2.pdf
Arquivo BibTex: CBRN1994-paper2.bib