Representing and Coping with Recurrent Change in Dynamic Constraint Satisfaction Problems

In many cases, constraint satisfaction problems (CSPs) are dynamic, so that changes can occur that invalidate the current solution. If such changes are temporary and recurrent, it may be possible to nd solutions that are less likely to be lost following a given change. (We call these stable solutions.) In this paper we compare the standard dynamic CSP model that we have used in the past with a static probabilistic model; we show that the former is more adequate for representing recurrent change and better suited to searching for solutions that are also stable. Using the dynamic CSP model, we compare two approaches to nding stable solutions that estimate either the likelihood of change in speciic problem elements or the relation between assignments and solution loss, and show how viable techniques can be obtained in both cases (although the former were generally superior in the situations tested).