The basic bound on the generalisation error of a PAC learner makes the assumption that a consistent hypothesis exists. This makes it appropriate to apply the method only in the case where we have a guarantee that a consistent hypothesis can be found, something that is rarely possible in real applications. The same problem arises if we decide not to use a hypothesis unless its error is below a prespecified number.
|Conference||16th Annual Conference on Learning Theory and 7th Kernel Workshop, COLT/Kernel 2003|
|Period||2003/08/24 → 2003/08/27|