From: predictr@bellatlantic.net   
      
   russell kym horsell wrote:   
   > Quite a few years ago some group tried to analyze some photos.   
   > In the typical manner of the day, the photos were meant to be classified   
   > as "containing a hidden military asset" or "not containing".   
   >   
   > A set of photos containing said assets (partly hidden behind  vegetation   
   > or cam netting, etc) was data-mined for statistically significant features,   
   > and an "average" model in feature-space extracted. A similar process   
   > was used to extract an "average" not-a-asset dataset.   
   > The idea was that a utility function was to be used -- based on   
   > some vector norm/distance type thing -- to decide whether any new photo   
   > (after relevant feature extraction) was "closer" to "has asset" than   
   > "does not have asset".   
   > Surprisingly, the 2 sets could be separated by the hyperplane and the   
   > automaton created was beleived to be very reliable.   
   >   
   > However, the customer brought along some new photos, none of which   
   > could be correctly classified by said automaton.   
   >   
   > It later trurned out the training phots had an unusual feature.   
   > Most of the phots of the "with asset" had been taken on a particular sunny   
   day;   
   > those "whithout asset" had been taken on overcast days.   
      
   Can you provide any details on this?  What research group?  What   
   customer?  When it happened?  Any references which are specific as to   
   such details?   
      
      
   -Will Dwinnell   
   http://will.dwinnell.com   
      
   [ 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)