> I do not uderstand why would you have multiple layers, after all every   
   > layer predicts corresponding values in their preceding ones, so it   
   > feels like they all should converge to near-input values?   
      
   That may be true with the "Rolls Royce" system - but if there is a   
   competition element in the learning, as used in the line detection   
   system then nodes in the output layer become detectors of higher   
   level features of the input layer. Adding extra layers should produce   
   detectors of successively higher level features.   
   My system is (almost) logically equivalent to HTM so should have all   
   the same   
   characteristics, only with greater biological plausibility.   
      
   Mick.   
      
   [ comp.ai is moderated ... your article may take a while to appear. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)