1
$\begingroup$

The statistical partition function (from Boltzmann I think) looks like

$\sum_\mathbf{u} e^{-E(\mathbf{u})}$

It's used in Boltzmann machines, for example:

$P(\mathbf{v}) = e^{-E(\mathbf{v})}/\sum_\mathbf{u} e^{-E(\mathbf{u})}$

Say I want to calculate the log, base e of the partition function:

$log_e(\sum_\mathbf{u} e^{-E(\mathbf{u})})$

but I don't want to calculate the energy of each state explicitly, because that would take too long with hundreds or thousands of nodes in a Boltzmann machine. The energy is given by the weights and states of the network:

$E(\mathbf{v}) = -\sum_i s^\mathbf{v}_i b_i -\sum_{i

How can I efficiently do this without calculating each state explicitly?

I think it has something to do with this, but I'm not quite connecting the dots yet.

0 Answers 0