This question is based on Honglak Lee's paper "Convolutional Deep Belief Networks for Scalable Unsupervised Learning of Hierarchical Representations".
I have implemented a convolutional RBM with probabilistic max-pooling based on equations in section 3.3. I now want to stack several of them as a convolutional DBN. I thought that this would simple, just connecting pooling layer to the next visible layer as a conventional DBN. But in the paper, there is a chapter "3.6 Hierarchical probabilistic inference". This chapter contains different formula for h^k and p^k. I don't get when these formula needs to be used and what is this chapter about ?
Can someone clarify this ?