What are the gold standard structure-learning methods here? I don't need anything particularly fast (my networks are fairly small). I'm aware of the structural EM algorithm (pdf) but it seems that the user needs to supply the algorithm with the number of hidden variables a priori. I've also read Elidan and Freidman's information bottleneck paper (pdf) which seems to do what I need, but, at first glance, appears complicated to implement (not that that's a deal-breaker, but simpler is better).
I'm mainly wondering if there is any recent research in this area that's simplified matters.
Thanks!
[link][comment]
