Abstract: The application of expected utility theory to construct adaptiveagents is both computationally intractable and statisticallyquestionable. To overcome these difficulties, agents need theability to delay the choice of the optimal policy to a later stagewhen they have learned more about the environment. Howshould agents do this optimally? An information-theoretic answerto this question is given by the Bayesian control rule – the solutionto the adaptive coding problem when there are not only observationsbut also actions. We review the central ideas behind the Bayesiancontrol rule.
- This event has passed.