From: mailbox@dmitry-kazakov.de   
      
   On 5 Oct 2003 19:56:06 -0700, wsiler@aol.com (William Siler) wrote:   
      
   >Dmitry A. Kazakov  wrote in message    
   ews:...   
   >> On 1 Oct 2003 11:54:53 -0700, abos@br.inter.net (Bruno) wrote:   
   >>   
   >> >Nobody has???   
   >>   
   >> Neural networks provide a way to represent and organize data. Fuzzy   
   >> logic is an extension of the conventional logic. What is to compare   
   >> here? Absolutely nothing! A valid comparison could be neural vs. fuzzy   
   >> neural networks. But it would be just crisp vs. fuzzy.   
   >>   
   >Actually, neural networks can provide a large number of functions,   
   >including (for example) classification. Thus for certain problems one   
   >can use either a neural net or a fuzzy expert system. They are, of   
   >course, quite different techniques.   
      
   I would disagree. A fuzzy expert system could be built on the basis of   
   a neuronal network. I mean, as the knowledge carrier one could take a   
   network instead of a data base of rules. From this point of view a   
   valid comparison could be: rules data base vs. neuronal network. And   
   again the word "fuzzy" have slipped away! (:-))   
      
   >An advantage of neural nets is that little or no a priori knowledge is   
   >required;   
      
   This only means that the learning algorithm is not tunable. It has no   
   parameters. Whether it is an advantage, is another question.   
      
   > the corresponding disadvantage is that after the neural net   
   >is constructed and tuned, one has little or no idea how it reaches its   
   >conclusion. There has been a lot of work into extracting a rule base   
   >that corresponds to the neural net from the neural net connections,   
   >but I have the impression that the results of this work are not   
   >terribly satisfactory.   
      
   Me too.   
      
   >However, I am not up to date on this work, so   
   >what I say may not be correct.   
   >   
   >A disadvantage of neural nets is that a training set of data and the   
   >corresponding conclusions is abolutely required. Some applications   
   >cannot possible meet this requirement.   
      
   After all it is a learning with a teacher, so it has its   
   disadvantages, like a necessity to have that "teacher".   
      
   >Expert systems do not usually require a training set of data for their   
   >construction; instead, expert kowledge is used to construct the rules.   
      
   Yes, but it again, it is comparing apples and oranges in my view.   
      
   >They do, of course, require calibration of model parameters and   
   >subsequent validation on real-life data. Neural nets also require   
   >validation.   
   >   
   >This is a very short treatment of the subject, but is perhaps better   
   >than nothing. I know of no publication that deals with the topic.   
      
   ---   
   Regards,   
   Dmitry Kazakov   
   www.dmitry-kazakov.de   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)