DBM pretraining issue

[STZhang]

I am really interested in your work 'Multimodal learning with deep boltzmann machine' and I have read your deepnet code.
When learning the parameters of the first layer DBM, the bottom-up weights are constrained to be twice the top-down weights. But I have trouble finding the parameter that related to "twice" or the function to copy the input nodes. Would you mind helping me with this?
Sorry for the trouble and Thanks a lot for your help.

[SGK]

edge up_factor might be what you're looking for.

[Steven Hoffman]

@STZhang , were you able to create a script to train a multimodal DBM, such as something similar to the runall_dbn.sh script used to train and test the multimodal DBN example? If so, would you be able/willing to provide the script you used or describe how to best create one?

Thank you in advance for any help you can provide.