We have a loose argument of how to find epsilon star. Perhaps we can improve it a bit with a more concrete way to estimate the KL between DGP and P(\xi | \hat{\eta}). Some thoughts are to use the bregman disvergence if that's easier to approximate, check the KL formulation in the paper that helped us get Lemma 1 and arguments of the type if the KL is bounded... etc.
We have a loose argument of how to find epsilon star. Perhaps we can improve it a bit with a more concrete way to estimate the KL between DGP and P(\xi | \hat{\eta}). Some thoughts are to use the bregman disvergence if that's easier to approximate, check the KL formulation in the paper that helped us get Lemma 1 and arguments of the type if the KL is bounded... etc.