It’s the frequentist vs. bayesian debate again. Frequentist methods, like worst case analyses, give hard guarantees no matter what the probability distribution over the environment is. Bayesian methods, like average case analyses, require a probability distribution over the environment and may mess up horribly if that distribution is wrong.
It’s the frequentist vs. bayesian debate again. Frequentist methods, like worst case analyses, give hard guarantees no matter what the probability distribution over the environment is. Bayesian methods, like average case analyses, require a probability distribution over the environment and may mess up horribly if that distribution is wrong.