From: wsilvert@netcabo.pt   
      
   This sounds like a project I have been working on. We developed a fuzzy   
   procedure to classify the benthic impacts of fish farming (Angel et al.   
   1998) and then used a neural network to analyse the evaluations of the   
   experts to create a fuzzy expert system (Silvert and Baptist 2000). The   
   procedure was to provide a team of experts with data, they evaluated the   
   data and produced the classification - we then fed the original data and the   
   expert conclusions into an NN to see if we could construct a black box that   
   would generate the same classification scheme.   
      
   Unfortunately we did not really have enough samples to construct a reliable   
   NN, although the results were promising. This is a common problem with the   
   use of NN in ecological situations, but I think the basic idea of feeding   
   both raw data and expert evaluations of those data into a neural network is   
   promising.   
      
   References below. Bill Silvert   
      
   Dror Angel, Peter Krost and William Silvert. 1998. Describing benthic   
   impacts of fish farming with fuzzy sets: theoretical background and   
   analytical methods. J. Appl. Ichthyol. 14: 1-8.   
      
   William Silvert and Martin Baptist. 2000. Can Neuronal Networks be used in   
   Data-Poor Situations? Presented at Int. Workshop on Applications of   
   Artificial Neural Networks to Ecological Modelling, Toulouse, France, 14-17   
   Dec. 1998. In: S. Lek & J.-F. Guégan (Eds.), Artificial Neuronal Networks;   
   Application to Ecology and Evolution. Springer-Verlag, Berlin, p. 241-248.   
      
      
   "William Siler"  wrote in message   
   news:49b9df3d.0310170616.473e9782@posting.google.com...   
   > Dmitry A. Kazakov  wrote in message   
   news:<00hsovgdr25m4jj69tgve8brhj8s5b6d4d@4ax.com>...   
   >   
   > > I would not set off NN against expert systems. Because it is   
   > > theoretically possible to build an expert system on the basis of a NN.   
   > > Consider "experts", which knowledge can be extracted in the form a   
   > > neuronal subnetwork. Then this knowledge, a description of a network,   
   > > is incorporated into a larger system using some standard framework.   
   > > That would be a NN-based expert system.   
   >   
   > This is a new idea to me, and is very interesting. The idea of looking   
   > at a neural net that has been conventionally constructed (from a   
   > training data set) as a fuzzy expert system, and extracting the rules   
   > from the resulting neural net, is of course not new; we seem to agree   
   > that this work has not been very satisfactory. But the idea of looking   
   > at a fuzzy expert system that has been constructed from expert   
   > knowledge rather than a training data set is new to me, and strikes me   
   > as being fairly important.   
   >   
   > Of course, real-world fuzzy expert systems are a lot more complicated   
   > than simple one-step fuzzy control expert systems. The expert systems   
   > I have constructed ten to be multi-step affairs, with the rules   
   > fireable in one step being fired in parallel, and the results of one   
   > step being fed as input to the next step. I think you mean that we   
   > could look at each epert system step as a layer in a neural net.   
   >   
   > We frequently have recursion, with the outputs being fed back as   
   > inputs to steps that have already been fired. Has anyone worked with   
   > neural nets in which the outputs of one layer are fed back as inputs   
   > to the same or a preceding layer?   
   >   
   > Sincerely, William Siler   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)