Quote:
Originally Posted by Willem
In lecture 2 it is mentioned that noisy targets are the sum of a deterministic target and noise, and the expected value of y given x is the true deterministic target function. I was wondering if we may always assume that the expectation value of the noise is zero. Yes for Gaussian noise, but is it true for other types of noise, perhaps systematic effects?

This is actually in Lecture 4.
The expected value of the noise in those cases would be a deterministic function of
and therefore can be taken as part of the target function itself, leaving the noise part with zero mean. We simply define the "noise" to be
, which forces the noise to have zero mean since the mean is taken out. There are other ways to define the noise, but they do not change the essence of the problem.