Mixed Constraint Satisfaction: A Framework for Decision Problems under Incomplete Knowledge

No. 1: Agents, AI in Art and Entertainment, Knowledge Representation, and Learning

Volume

Proceedings of the AAAI Conference on Artificial Intelligence, 13

Constraint Satisfaction

Constraint satisfaction is a powerful tool for representing and solving decision problems with complete knowledge about the world. We extend the CSP framework so as to represent decision problems under incomplete knowledge. The basis of the extension consists in a distinction between controllable and uncontrollable variables - hence the terminology "mixed CSP" - and a "solution" gives actually a conditional decision. We study the complexity of deciding the consistency of a mixed CSP. As the problem is generally intractable, we propose an algorithm for finding an approximate solution.