I am designing RBM with CD training methodology in MATLAB. CD is an unsupervised learning algorithm, so I do not need to use any label information included in input. Does that make sense?
Restricted Bolzmann Machine is one of the unsupervised learning method that finds latent feature for the input. If you want to use it as supervised way, you must add another softmax layer to the top layer of RBM in case of 1-of-K classification. You can do with CD as well.