utility

The "worth" in a decision

Need to unify with the algorithms book, but the wikipedia definition seems pretty intuitive; the utility $U$ corresponds to the expectation value of the reward $x$ , with probabilities $p$ :

$U(p) = \sum p_k x_k$

...and "rational decisions" are basically those that maximize the expected reward.

#needs-expanding

Backlinks

markov-decision-process

- > ...treated as components in an additively decomposed utility function. In a finite horizon problem with $n$ decisions, the [[utility]] associated with a sequence of rewards $r_{1:n}$ is given by $\sum_{t=1}^{n} r_t$

decision-theory

- A *utility* node represents a [[utility]] variable, noted by diamonds and cannot have children

policy

The definition of an optimal policy $\pi*$ is one that maximizes the [[utility]]: