![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
guslTonight, Percy Liang presented an interesting poster on asymptotically optimal regularization: for a certain class of regularizers, they show how to minimize risk asymptotically.
I was somewhat surprised to learn that this optimal estimation can be interpreted as MAP estimation with a two-level hierarchical prior.
But perhaps my surprise is surprising. I seem to be behaving as if there are no procedures with good frequentist properties that can't be interpreted as Bayesian inference for some prior!
[I wish to leave it ambiguous whether this is sarcastic... as I in fact do not know what to think]
I was somewhat surprised to learn that this optimal estimation can be interpreted as MAP estimation with a two-level hierarchical prior.
But perhaps my surprise is surprising. I seem to be behaving as if there are no procedures with good frequentist properties that can't be interpreted as Bayesian inference for some prior!
[I wish to leave it ambiguous whether this is sarcastic... as I in fact do not know what to think]