From: earlcox@earlcoxreports.com   
      
   Bill,   
      
     a better way to measure the stability as well as the validity of a fuzzy   
   model is to track the amount of evidence underlying the outcome fuzzy set of   
   each solution variable (see the discussion of this exact issue in my Fuzzy   
   Systems Handbook (2nd Ed)). By tracking the slop of a regression line   
   through the degree of evidence in the outcomes (adjusted for statistical   
   noise) you can detect several kinds of performance instabilities.   
      
    Earl   
      
      
   "William Siler"  wrote in message   
   news:49b9df3d.0401071936.48db590c@posting.google.com...   
   > "Bartosz Bien"  wrote in message   
   news:...   
   > > As William has already stated, I'm attempting to model such problems   
   with   
   > > cascaded fuzzy components. Specifically, they are Mamdani FLCs of quite   
   > > simple structure (which make processing faster), and I was having them   
   in   
   > > mind when I wrote the reply for tarmat.   
   > >   
   > Your problem is a nice one. Without any experience, let me make a   
   > tentative suggestion.   
   >   
   > I have written a number of fuzzy expert systems for event detection   
   > and classification. It occurs to me that detecting instability might   
   > be a fairly easy problem for a rule-based fuzzy expert system. If the   
   > expert system could be hooked to the neural net, with the outputs of   
   > the "neurons" as input to the expert system, it might not be too   
   > difficult to detect abnormal conditions and adjust the neural net   
   > parameters to correct them by adjusting weights as expert system   
   > outputs.   
   >   
   > Again, I have no experience with hybrid systems, so this idea is put   
   > forth quite tentatively.   
   >   
   > William Siler   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)