Title:	Dealing with inconsistencies and knowledge loss in combinatioral neural model.
Authors:	PRADO, H. A. do ENGEL, P. M. SILVA, K. C. da
Affiliation:	HERCULES ANTONIO DO PRADO, CPAC; PAULO MARTINS ENGEL, UNIVERSIDADE FEDERAL DO RIO GRANDE DO SUL; KATIA CILENE DA SILVA, UNIVERSIDADE FEDERAL DO RIO GRANDE DO SUL.
Date Issued:	2001
Citation:	In: SIMPOSIO ARGENTINO EN INTELIGENCIA ARTIFICIAL - ASAI'2001; JORNADAS ARGENTINAS DE INFORMATICA E INVESTIGACION OPERATIVA, 30., 2001, Buenos Aires. Anales JAIIO. Buenos Aires: Sociedad Argentina de Informatica e Investigacion Operativa, 2001.
Pages:	p. 74-84.
Description:	During the last years much effort has been devoted to generate knowledge through Data Mining techniques. Despite of all advances, few efforts have been addressed in cope with post processing problems. This paper is about those problems in Combinatorial Neural Model (CNM). CNM, a supervised learning algorithm introduced by Machado, received many improvements that made it useful for Data Mining. Two main problems arc approached. The first one corresponds 1o the conflicts that can emerge in the knowledge base since the model acquires knowledge from Different sources, cither specialists or examples. In this way, we apply the concept of extended negation, as the Boolean negation is not so natural. The second problem arises after applying the pruning process to CNM. Since the model is incremental, any part of the knowledge base pruned after the training process in time t, can be important to the training process in time. Disregarding this pruned part can Lend to loss of knowledge. Considering that it is not possible to avoid pruning, and thus to maintain the knowledge base untouched during all its lifetime, we propose an approach to mitigate the problem.
Thesagro:	Base de Dados Programa de Computador Tecnologia da Informação Informática
Keywords:	Inteligencia artificial Mineração de dados Redes neurais
Type of Material:	Artigo em anais e proceedings
Access:	openAccess
Appears in Collections:	Artigo em anais de congresso (CPAC)