From: ted.dunning@gmail.com   
      
   I think that transduction is generally better viewed from a Bayesian   
   framework as a missing data problem.  This is pretty blasphemous to   
   many Vapnitchiks, but it makes things much easier to think about and it   
   certainly makes it easier to find related literature on the topic since   
   the word transduction is not so widely used as it might be to refer to   
   partially supervised training.   
      
   When regression is looked at this way, it is pretty straightforward to   
   see how to proceed and pretty simple to put some quick bounds on how   
   much it will help since knowing the distribution of the independent or   
   dependent variables isn't nearly as important as knowing the   
   distribution of the error process.  In particular, the question of   
   whether the error process has heavy tails is key (i.e. whether outliers   
   should be rejected).   
      
   [ comp.ai is moderated.  To submit, just post and be patient, or if ]   
   [ that fails mail your article to , and ]   
   [ ask your news administrator to fix the problems with your system. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)