Fuzzy Neural Net

EditHint: merge with ArtificialNeuralNetwork.

ArtificialNeuralNetworks are an effective tool for classification of patterns (well, vectors). However, one disadvantage of neural net modeling is that the knowledge acquired by neural nets is hard to extract. FuzzyNeuralNetwork?s can use BackPropagation? to identify and express input-output relationships in the form of fuzzy rules. This leads to the possibility of human knowledge extraction.

Could you explain what it is about fuzziness that enables FNNs to "use back-propagation to identify and express input-output relationships in the form of fuzzy rules"? Why can't you do basically the same thing (whatever exactly it is) with ordinary neural networks?

---

Formerly stated as: A NeuralNet? that uses FuzzyLogic weightings instead of discrete number weightings.

A neural net does not necessarily use discrete numbered weightings unless it is on a finite TuringMachine. Similarly with FNNs. This is because it is impossible to represent a number with infinite precision on a finite TuringMachine. The "finite" refers to the size of the store.

---

No online reference known to Wikizens. Some are available in textbooks...

I read about one that was used to control a helicopter in high winds. In a book on FuzzyLogic, I can't remember which.

Found this ref online at http://www.cis.uab.edu/info/faculty/reilly/96HerraraVergegay.html

 "James J. Buckley, Kevin D. Reilly, Krishnamraju V. Penmetcha 
 Backpropagation and Genetic Algorithms
   for 
 Training Fuzzy Neural Nets 
 F. Herrera and J. L. Verdegay, Eds. 
 Genetic Algorithms and Soft Computing. 
 Phisyca-Verlag Studies in Fuzziness, pp. 505-532. 

 Abstract

"This chapter concerns combined backpropagation and genetic training of fuzzy neural nets whose weights and signals are given as real or triangular fuzzy numbers. The proposed fuzzy neural network with backpropagation and genetic-based learning system is used on problems which map a fuzzy or real input to a fuzzy or real output based on interval arithmetic operations. Experimental results demonstrating characteristics of various non-linear mappings are discussed. "

---

Can't talk about Fuzzy anything without mentioning its hero Bart Kosko . His books "Fuzzy Logic" and "Fuzzy Engineering" are great. The latter discusses FNN's and adaptive non-linear mapping.

It was while browsing FuzzyThinking by BartKosko? that I came up with the idea of a FuzzyCommunity. -- PaulMillar

Would you think that a FuzzyNeuralNet could run over something like PondOpenNetworkDistributed? -- MichaelKing

---

CategoryFuzzy