Construction of a Novel Neuro-Fuzzy Network

摘要

This paper presents a novel neuro-fuzzy network structure -the Mega-Cell Neural Network. This network is motivated by an on-going research on the nature of fuzzy rule binding. Its major differences from other neuro-fuzzy networks are: (1) all nodes in the network are virtually the same; (2) membership values in this network become emergent properties; (3) this neuro-fuzzy network embodies the notion of distributive rule binding. Cn the other hand, this neural network distinguishes inputs from different variable, whereas in popular neural networks, the distinction is not obvious. This neural network builds around the neuron of Mega-Cell. The Mega-Cell also differs from traditional neurons in that (1) it accommodates both input and output credits; (2) inputs from different variables into this neuron are T-normed together to produce internal firing degree (IFD). Moreover, this neuron is fuzzy in nature. Apart from detailed descriptions on the operating mechanisms of Mega-Cell and Mega-Cell Neural Network, a brief analysis is also provided in this paper. Some notes on biological bearings of Mega-Cell come in at the Appendix.