From: ted.dunning@gmail.com   
      
   In some sense, the hypothesis was encoded in bits.  Think about where   
   you heard about it.   
      
   In a stronger sense, however, it is much more useful mathematically to   
   view minimum description length techniques as maximum posterior   
   likelihood estimators.  As such, they are regularized versions of   
   maximum likelihood estimators.  These have advantages in that certain   
   singular conditions can be avoided (mixtures of multi-variate   
   Gaussians are a class example), but they have the common problems of   
   all methods that provide a single estimation of the model as opposed   
   to estimating the posterior distribution over model parameters (which   
   allows better inference).   
      
   On Aug 23, 5:15 am, David  wrote:   
   > Hi,   
   >   
   > I understand that MDLP states the hypothesis that minimizes the sum of   
   > its description length and the description length of data given the   
   > hypothesis, is selected. This is how MDLP is defined in "Machine   
   > Learning" by Tom Mitchell 1997. Binary codes are used to represent the   
   > probabilities of hypotheses. The length is the number of bits to   
   > represent these probablities of the hypotheses.   
   >   
   > Q: Why isn't the hypothesis itself represented in binary codes?   
   >   
   > Many thanks,   
   >   
   > David   
   >   
      
   [ 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)