Keywords:

Abstract:

k- and t-optimality algorithms [9, 6] provide solutions to
DCOPs that are optimal in regions characterized by its size
and distance respectively. Moreover, they provide quality
guarantees on their solutions. Here we generalise the k- and
t-optimal framework to introduce C-optimality, a
exible framework that provides reward-independent quality guar-
antees for optima in regions characterised by any arbitrary
criterion. Therefore, C-optimality allows us to explore the
space of criteria (beyond size and distance) looking for those
that lead to better solution qualities. We benet from this
larger space of criteria to propose a new criterion, the so-
called size-bounded-distance criterion, which outperforms k-
and t-optimality.

• Tagged
• XML
• BibTex