[Topic-models] LDA paper by Blei - some more questions

David Blei blei at CS.Princeton.EDU
Wed Nov 19 08:44:15 EST 2008

hi mateusz,

in the E-step, optimize variational parameters for each document and  
compute expected sufficient statistics under the resulting variational  
distributions.  this step depends on a setting of the model  
parameters, i.e.,  the dirichlet hyperparameters and topic  
distributions.  in the M-step, optimize those model parameters  
according to those expected sufficient statistics.  repeat until the  
bound on the log likelihood of the data converges.


On Nov 14, 2008, at 4:45 PM, Mateusz Berezecki wrote:

> So the first step of the algorithm explicitly relies on the inference
> phase being completed as the variational parameters are valid
> (optimized) only after inference phase (?).

More information about the Topic-models mailing list