CODEWITHSUNDEEP
neural-networkdeep-learningrbmdbn
Yoko
Yoko

Reputation: 11

Deep Belief Network inference: hidden layers need random number generator?

I am learning Deep Belief Network and Restricted Boltzmann Machine.

  1. In training DBN (CD-1, greedy, layer-wise), inputs to the second, third, and nth RBM should be stochastic binary (0 or 1) and not probability?

  2. As for the inference process in DBN, are hidden units also stochastic binary and not probability? Can sigmd{Σ(W*V+b)} be used as input to the layer immediately above? Or do I further need a random number generator to obtain stochastic binary result for h units then use these h values as inputs to the layer right above?

Could someone please explain to me?

Upvotes: 1

Views: 170

Answers (0)

Related Questions