Quote:
Originally Posted by OlivierB
What does "target distribution" mean exactly in the context of Q4 ? All other definitions are clear to me from the lectures & book, except this one.
In my understanding the target function is determined, fixed, and it has no 'probabilistic aspect', as maybe the 'distribution' term suggests.
I would like to know what 'Stochastic noise does not depend on the target distribution' means in detail ?

The target distribution describes the situation with noisy targets, where it is no longer a function but a function plus noise that is described by a distribution
. You can look at slide 16 of Lecture 4 for details (in the Machine Learning Video Library, the relevant segment is
http://work.caltech.edu/library/044.html).