| Define
input vector that contains input patterns |
| with −1 bias |
| For to iter |
| Calculate the hidden neuron output according
to where |
| the net weighted input is falling in the range of the |
| piecewise sigmoid-logarithmic function |
| Define output vector from the hidden layer
that |
| contains hidden neuron output and −1 bias |
| Calculate the final neuron output, the
first back- |
| propagation error set, and the second back-propagation |
| error set according
to where the net weighted input is |
| falling in the range of the piecewise
sigmoid-logarithmic |
| function |
| Check the criterion of percentage of the
input data |
| that has an
error less than 20% |
| If all input data have errors less than
20%, |
| stop the training |
| Update the second weight matrix |
| Update the first weight matrix |
| End |