Suppose that I have a model M that overfits a large dataset S such that the test error is 30%. Does that mean that there will always exist a model that is smaller and less complex than M that will have a test error less than 30% on S (and does not overfit S).
Asked
Active
Viewed 98 times
