Hello All,
I am confused about the definition of covering number. I think we need some points x1, ..., xm first to have an epsilon approximation of F, just like the S_F(x1,...,xm). And then is the covering number, referred to in the question for a particular set of points or maximum over all such point like the growth function S_F(m)?
I am not clear what is meant by "epsilon approximation holds uniformly on X".
Also do we need to find a bound on fat shattering VC dimension by the covering number for same epsilon? I mean can one epsilon be a multiple of other
Thanking you,
Manzil