文档介绍：IEEE TRANSACTIONS ON WORKS, VOL. 10, NO. 4, JULY 1999 801
Unification of Neural and Wavelet
Networks and Fuzzy Systems
Leonardo M. Reyneri, Member, IEEE
Abstract— This paper analyzes monly used soft product implicators and those with rules containing only a
computing paradigms (neural and works and fuzzy partial number of inputs.
systems, Bayesian classifiers, fuzzy partitions, etc.) and tries to Later, Hunt et al. [10] further extended the functional equiv-
outline similarities and differences among each other. These are
exploited to produce the weighted radial basis functions paradigm alence to Takagi–Sugeno consequents [11], but this required a
which may act as a neuro-fuzzy unification paradigm. Training modification of the original RBF paradigm.
rules (both supervised and unsupervised) are also unified by the The equivalence of weighted radial basis functions [8]
proposed algorithm. Analyzing differences and similarities among with perceptrons and a preliminary attempt to neuro-fuzzy
existing paradigms helps to understand that many puting
paradigms are very similar to each other and can be grouped in unification have been first described in [9], although with some
just two major classes. The many reasons to unify puting limitations, especially in the unification of perceptrons, on the
paradigms are also shown in the paper. A conversion method is one hand, and RBF’s and FS’s, on the other hand.
presented to convert perceptrons, radial basis functions, wavelet Another work from Benitez et al. [12] has proposed an inter-
networks, and fuzzy systems from each other. esting approach to the functional equivalence of perceptrons
Index Terms—Artificial works, function approxima- and FS’s, mainly for the aim of interpreting the knowledge
tion, classifiers, fuzzy systems, learning rules, taxonomy. hidden within perceptrons. The work introduces an ad hoc
inference operator (the interactive-or), which is the bridge
I. INTRODUCTION between perceptrons and