An interesting fact about the Viterbi algorithm is that, from reading the paper introducing it, it seems that Viterbi did not realize that it actually produced the exactly-correct answer. He thought of it as some sort of approximation.

]]>As you mentioned in the post:

“it may also be more efficient than a low energy Metropolis-Hastings algorithm”.

Actually, I did come across this kind of test results.

But, I very confused and curious about the theoretical basis for it being more efficient?

Could you help me about this?

Best.

Thanks.

Regarding nuisance parameters: On the one hand, Mayo’s more statistically informed collaborators like Cox and Spanos dismiss the issue of nuisance parameters rather breezily, pointing to those nice conditions you note. On the other hand, it would be fairly straightforward to apply severity reasoning in the context of more modern frequentist methods, some of which work around nuisance parameters, e.g., the bootstrap, and others of which work well even in their presence, e.g., higher order likelihood inference. So I’m mostly inclined to give Mayo a pass on that point.

Mayo’s attacks on likelihood fall short for me, as I describe here: https://itschancy.wordpress.com/2018/12/01/challenges-to-the-likelihood-principle-in-mayos-new-book/

I have some critical blog posts exploring the severity concept in the works.

]]>