InfraBayesianisn a theory of reinforcement learning in the nonrealizable setting based on imprecise probability by Vanessa Kosoy and Alex Appel. It is part of Vanessa Kosoy’s Learning-Theoretic Agenda in AI alignment.
Non-realizable in this context refers to the situation where the ‘true’ distribution does not necessarily lie in the model class of the (reinforcement) agent.
Infradistributions generalize the credal sets of imprecise probability which in turn generalize probability distributions.
Last revised on May 6, 2023 at 15:39:21. See the history of this page for a list of all contributions to it.