|
i want to make gaussian visible and binary hidden RBM and train using CD-1 method.Want to do in matlab.I am just learning to work in RBM. Donot have much background on this area. for that from already existing codes of hinton's binary visible and gaussian hidden RBM, i learnt we need to normalize the values using normal distribution. So, to make gaussian visible units and binary hidden units,i took the MNIST training dataset and normalzed as below data = batchdata(:,:,batch); data1 = data+ randn(size(data)); when i do this ,the trained data is more distorted..can anybody give clear idea about it is that make sense? |
