Convergence mechanism of learning on backpropagation and high speed learning

The input signals of units on neural network (NN) are converted to output signals by an I/O function such as the sigmoid functions. The input signals at each layer unit is changed by the learning. Its change is dependent on the slope of the I/O function. In the sensitive region (active region) given by the I/O function, input signals move effectively and the movement increase with the increase of the slope. At the saturation region, however, it is very small. The limited linear function is used instead of the sigmoid function, because the active region is defined clearly and an examination of the movement is made easily. The convergence mechanism of learning by back propagation is examined by the movement per training cycle. The possibility of the high speed learning is considered on the basis of the results