From: rodrigo.starr@poli.usp.br   
      
   On some kinds of NN it's quite straightforward to see a fuzzy system (I   
   mean, to understand what kind of knowledge is stored in each network   
   node). One kind of NN that have this property is the CMAC. The book   
   Neurofuzzy Adaptive Modelling and Control, by Brown and Harris, talks a   
   little about this problem.   
      
   Guillaume wrote:   
   >>> -- only that the comments in this thread have not focused on what   
   >>> fuzzy logic actually means and what fizzy logic actually does and how   
   >>> it actually works.   
   >>   
   >>   
   >> Maybe because nobody actually knows.   
   >   
   >   
   > A bit strong a statement. I'd rather say a lot of people don't care   
   > to figure it out. Big difference.   
   >   
   > I have seen a lot of people (including engineers) considering fuzzy   
   > logic either as a "black box" or as a gadget that actually may or may   
   > not work.   
   > Neither views are leading to any interesting results.   
   >   
   > Having said that, I might add my "grain of salt".   
   > I personally think that there is no fundamental difference between   
   > a trained neural network at a given point and a fuzzy logic system.   
   > Both work with imprecise input values and internal rules to yield   
   > specific results. A given trained NN is functionally equivalent to   
   > a fuzzy system. The big advantage of FL being, of course, that you   
   > can actually create the rules - whereas, in a NN, there is no or little   
   > way you can control or even figure out the internal rules as they are   
   > evolving during the training phase.   
   >   
   > An interesting research topic would be whether it is possible to   
   > transform a given NN into a FL system and vice versa, and in which   
   > cases. I don't know if it has already been studied.   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)