I agree that in practice PAC designates a specific subset of ideas and papers, my point is just that “probably approximately correct” doesn’t help to distinguish that subset (words like boosting and multi-armed bandit do). The VC-theory is totally based on PAC-style results (though it would be better to say PAC is based on VC-style results) and the MDL/MML people have similar generalization theorems as well.
I agree that in practice PAC designates a specific subset of ideas and papers, my point is just that “probably approximately correct” doesn’t help to distinguish that subset (words like boosting and multi-armed bandit do). The VC-theory is totally based on PAC-style results (though it would be better to say PAC is based on VC-style results) and the MDL/MML people have similar generalization theorems as well.
I agree. What are MDL/MML though?
Minimum description length/minimum message length.