Application of Exchanging Monte Carlo Method to Sample Deep Boltzmann Machines
Hiroki Shibata, Lieu-Hen Chen, Yasufumi Takama · 2020
There is a method to learn the parameters of Boltzmann Machines such that using contrastive divergence, or parallel tempering. However, none of them can be applied to multilayered BM in general way. To this problem, we introduce a new sampling method based on Exchanging Monte Carlo to estimate the distribution of BM with combining crystallizing effect of simulated annealing on Boltzmann machines.