3 Learning in Artificial Neural Networks
• The neuron must be trained to recognize the input patterns and classify them to give the corresponding output. The procedure is to present the sequence of the four input patterns to the neuron so that the weights are adjusted after each iteration (using feedback of the error found by comparing the estimate to the desired result). This step is repeated until the weights converge to a uniform set of values that allows the neuron to classify each of the four inputs correctly. The results shown in Table 6.1 were produced in Excel. In this simple, a threshold function is used to evaluate the summation of input and the desired values is used to update the weights, subsequently reinforcing the output and the desired values is used to update the weights, subsequently reinforcing the correct results. At any step in the process, for a neuron j we have: