Modifying a existing RBM based deep autoencoder according to a research paper



Hi !!!

This is Aman here. I am working on this paper which tells about how to work on short sentences and do topic modelling. I have looked at this GitHub repo where RBM based encoder is implemented using Tensorflow. But as you can see in the paper that the author has used a customised RBM by adding a few more term called penalty term.

I am doing a mini-batch learning here, so I am struggling to modify the code in order to make the model work on that customised equation. If someone can help me how to modify the same.

Aman Khandelia