We know how DELTA rule works from the previous presentation. Neuron that uses this rule to learn has following inner schematic:
The process of learning is described by following equation:
W’ = W + ηδX
whereW - previous weights vector
W’ - new weights vector
η - learn-rate parameter
δ - error of output signal compared to desired value
X - input vector
In this case weights vectors and input vector are normalized. Situation like this is presented in left column of graphs - click on button named Project. Our project shows how important is to normalize values in process of learning. Right column of graphs presents situation when non of vectors is normalized (please make note of scales of axes on the graph!). Neuron without normalization may not learn but, for example, oscillate and increase its lenght.