Uncertainty is a common feature of real-life combinatorial optimisation
problems. Constraint programming is known as a powerful tool for
tackling these problems, but current approaches to uncertain data can
lead us to solve the wrong problem because of the approximations made.
Such an outcome is of little help to a user who expects the right
problem to be tackled and reliable information returned.
In this talk we present the certainty closure framework for reliable
constraint reasoning in the presence of uncertain data. We explain how
to provide the user with reliable insight by: (1) enclosing the
uncertainty using what is known for sure about the data, and (2)
deriving a closure, a set of possible solutions to the uncertain
constraint problem. We explain the formal basis of our framework, and
illustrate the benefits on two case studies.
This is joint work with
Carmen Gervet.
|