One might guess that the backpropagation learning rules should be inversely related to $f^{\text{'}}(net)$

that is$,$ that weight change should be large where the output does not vary. In fact, it can

be observed from $w_{kj}=(t_k-z_k)f^{\text{'}}(ne{t}_k)y_j$ that

the learning rule is linear in $f^{\text{'}}(net).$ Explain intuitively why the learning rule should be linear in

$f^{\text{'}}(net).$