Parallel learning for back-propagation network in binary field

The major problems in training a backpropagation neural network, especially with the input and output sets ∈ {0, 1}ⁿ, are the slow speed and unknown number of neurons and unknown number of hidden layers. Two efficient techniques are proposed which are used in different situations to speed up the learning process. A parallel dynamic learning concept is introduced. Experimental results show that the learning speed is more than several hundred times faster than the regular training using the structure based on Kolmogorov´s theorem