Page 74:

When there is noise in the data, , where . If is a zero-mean noise random variable with variance , show that the bias-variance decomposition becomes

We have:

We split the above expression in two sub-expressions, in which:

Together, we can derive that: