Pareto analysis is a broadly applicable method to
model and analyze tradeoffs in multi-objective optimization
problems. The set of Pareto optimal solutions is guaranteed
to contain the best solution for any arbitrary cost function or
selection procedure. This work introduces a method to explicitly
take uncertainty into account during Pareto analysis. A solution
is not modeled by a single point in the solution space, but
rather by a set of such points. This is useful in settings with
much uncertainty, such as during model-based design space
exploration for embedded systems. A bounding-box abstraction is
introduced as a finite representation of Pareto optimal solutions
under uncertainty. It is shown that the set of Pareto optimal
solutions in the proposed approach still captures exactly the
potentially best solutions for any cost function as well as any
way of reducing the amount of uncertainty. During model-based
design space exploration, for instance, design and implementation
choices that are made during the development process reduce
the amount of uncertainty. Steps in such a refinement trajectory
can render previously Pareto optimal solutions suboptimal. The
presented results provide a way to ensure that early selections
in the refinement process remain valid.

Name | ES reports |
---|

Volume | 2011-01 |
---|

ISSN (Print) | 1574-9517 |
---|