Programming thread

Has anyone here worked with a repeating Boltzmann machine?  Can anyone share how it programmatically works?

>>1
hmm, yes a little. I like them too =) nearly as simple as a clustering algorithm, but it's a neural net..

So, lets see now..
They use a nifty forward/back/forward training method, which means the network also works as an auto-encoder at each layer..

Training is (roughly) as follows..

L^0a * weights -> L^1a
L^1a * weights' -> L^0b
L^0b * weights -> L^1b

L^0a' * L^1a' -> w^{Delta +}
L^0b' * L^1b' -> w^{Delta -}