Active Perception, Dual Control, and Exploration › Quiz

The predictive observation probability P(o|b,a) that normalizes the posterior is the same quantity that drives information gain, so the belief-MDP recursion and the entropy-reduction definition share the same engine.

The belief-MDP removes uncertainty entirely, so information gain is always zero inside it.

The belief-MDP transition is deterministic without observations, so information gain only appears at the terminal step.

Information gain is defined only for the extrinsic reward, independent of the belief transition.