Sérgio M. Dias,  Luis E. Zárate,  Bruno M. Nogueira,  Newton J. Vieira. 

Since the knowledge extracted from Artificial Neural Networks is not easily comprehensible by humans, many attempts have been done in order to ex­tract knowledge from these nets. In this paper, it is explored the FCANN method, which uses Formal Concept Analysis to extract and represent interpretable kno­wledge from neural networks. However, rule extraction in this method presents a high computational cost and, many times, generates rules that are hard to understand. So, this paper proposes an implication rules extraction from con­ceptual lattices build from frequent itemsets, increasing the method scalability and rules comprehensibility. This new approach was applied to a termosiphon system, obtaining good results.

http://www.lbd.dcc.ufmg.br:8080/colecoes/waamd/2009/002.pdf

Processando consulta.