Quote:
Originally Posted by menon.rx
I agree. However, the amount of noise depends upon N and the relative complexity of the target function. Therefore, isn't it true that one cannot make a general statement about trend without knowledge of the relative complexity between the hypothesis and the target function as well as N.

If you have the hypothesis set and and the target function defined (fixed), the deterministic noise should also be defined ("fixed") and independent of the amount of training data (N). The overfitting and Eout are dependent of N.