A Constraint Satisfaction Approach to Decision under Uncertainty
This work addresses decision-making under uncertainty for AI and operations research, but it appears incremental as it builds on the existing CSP framework.
The paper tackles the problem of extending the Constraint Satisfaction Problem (CSP) framework to handle decision-making under uncertainty by differentiating between controllable decision variables and uncontrollable parameters with probabilistic uncertainty, resulting in two algorithms for computing decisions with maximal probability and conditional decisions mapping cases to decisions.
The Constraint Satisfaction Problem (CSP) framework offers a simple and sound basis for representing and solving simple decision problems, without uncertainty. This paper is devoted to an extension of the CSP framework enabling us to deal with some decisions problems under uncertainty. This extension relies on a differentiation between the agent-controllable decision variables and the uncontrollable parameters whose values depend on the occurrence of uncertain events. The uncertainty on the values of the parameters is assumed to be given under the form of a probability distribution. Two algorithms are given, for computing respectively decisions solving the problem with a maximal probability, and conditional decisions mapping the largest possible amount of possible cases to actual decisions.