108
W. Xinying et al.
Fig. 10.2 RBM training process
Fig. 10.3 Typical deep network
in the part close to the visible layer, and the part farthest from the visible layer. Still
using RBM, you can get a deep confidence network.
10.2.3 Deep Reliability Network
The deep confidence network is a Bayesian probability generation model consisting
of multiple layers of random hidden variables. The upper two layers have undirected
symmetric connections, and the lower layers get directed connections from the upper
layer. As shown in Fig. 10.3, the basic structural unit of the DBM is an RBM, and
the number of visible layer nodes of each RBM unit is equal to the number of hidden
layer nodes of the previous RBM unit.
In the deep confidence network framework, the top two layers constitute the
associative memory, and the connections between the other layers are determined
by the top-down generated weights. In the training process, the value of the visible
layer unit is first mapped to the hidden layer unit, and then the visible layer unit is