Maximum Expected Utility

1. Acting Probabilistic Graphical Models Decision Making MaximumExpectedUtility

2. Simple Decision Making A simple decision making situation D: • A set of possible actions Val(A)={a1,…,aK} • A set of states Val(X) = {x1,…,xN} • A distribution P(X | A) • A utility function U(X, A)

3. Expected Utility • Want to choose action a that maximizes the expected utility

4. f0 f1 m0 0 -7 U m1 0 5 m2 0 20 Simple Influence Diagram m2 m0 m1 0.2 0.5 0.3 Market Found

5. VQ VG VS More Complex Influence Diagram Difficulty Intelligence Study Grade Letter Job

6. f0 f1 m0 0 -7 U m1 0 5 m2 0 20 Information Edges Market m2 m0 m1 s1 s0 s2 0.2 0.5 0.3 0.3 0.1 m0 0.6 Survey 0.3 0.4 m1 0.3 0.4 0.5 m2 0.1 Found Decision rule  at action node A is a CPD: P(A | Parents(A))

7. Expected Utility with Information • Want to choose the decision rule A that maximizes the expected utility

8. U Finding MEU Decision Rules Market Survey Found

9. f0 f0 f1 f1 m0 s0 0 0 -1.25 -7 U m1 s1 0 0 5 1.15 m2 s2 0 0 20 2.1 Finding MEU Decision Rules Market m2 m0 m1 s1 s0 s2 0.2 0.5 0.3 Survey 0.3 0.1 m0 0.6 0.3 0.4 m1 0.3 0.4 0.5 m2 0.1 Found

10. More Generally

11. MEU Algorithm Summary • To compute MEU & optimize decision at A: • Treat A as random variable with arbitrary CPD • Introduce utility factor with scope PaU • Eliminate all variables except A, Z (A’s parents) to produce factor (A, Z) • For each z, set:

12. Decision Making under Uncertainty • MEU principle provides rigorous foundation • PGMs provide structured representation for probabilities, actions, and utilities • PGM inference methods (VE) can be used for • Finding the optimal strategy • Determining overall value of the decision situation • Efficient methods also exist for: • Multiple utility components • Multiple decisions

13. END END END