# Talk:Boltzmann machine

It would be great if an outline of proof for equation (5) is given and especially the explicit formulae for <.>_{data} and <.>_{model} will be very welcome.

Editor: Providing proofs may not be a good idea for an encyclopedia, unless the proof itself is an encyclopedic topic.

## review of Boltzmann Machine article

This is an excellent article pitched at just the right level for an encyclopedia.

In the section of 'Relationships to other models' I think it's worth mentioning the relationship to undirected graphical models, as well as MRFs. Much of the literature on undirected graphical models would be relevant to someone trying to understand Boltzmann machines. Similarly there is a large Statistics literature on log-linear models, which might be relevant.

In the section on 'Learning with hidden units', it is worth mentioning that this algorithm can be derived from the generalized EM algorithm (Dempster et al. 1977). It might also be nice to explicitly state that it is a *local* learning rule.

1. I have added a section on the relationship to graphical models and moved the stuff about how a stack of RBM's is not a BM to that section.

2. i have put in "locally available" when describing the learning rule.

3. I havent put in the relationship to EM becasue people often get confused by the negative phase and its just to cumbersome to disabuse them (in a brief article).