In deep neural network architecture:
Calculating how many bits are required to represent a fully connected neural network. Reducing the length of network by half, that is cutting half of the branches and optimizing the choice of the connected branches with SA Algorithm. Then the neutral network will be allowed to find the branch weights using stochastic gradient descent.
* Setting the number of branches free, learning them with SA algorithm.
* Setting the number of branches and layer free, Learning with SA algorithm.