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.
Links:
[1] http://www.iiia.csic.es/en/individual/meritxell-vinyals
[2] http://www.iiia.csic.es/en/node/4091
[3] http://www.iiia.csic.es/en/individual/jesus-cerquides
[4] http://www.iiia.csic.es/en/individual/juan-a-rodriguez-aguilar
[5] http://www.iiia.csic.es/en/node/4092
[6] http://www.iiia.csic.es/en/individual/milind-tambe
[7] http://www.iiia.csic.es/en/node/4093
[8] http://www.iiia.csic.es/en/publications/keyword/DCOP
[9] http://www.iiia.csic.es/en/publications/keyword/approximate algorithm
[10] http://www.iiia.csic.es/en/publications/keyword/bound
[11] http://www.iiia.csic.es/en/publications/keyword/region optimality
[12] http://www.iiia.csic.es/en/publications/export/tagged/4098
[13] http://www.iiia.csic.es/en/publications/export/xml/4098
[14] http://www.iiia.csic.es/en/publications/export/bib/4098
[15] http://www.iiia.csic.es/en/project/at
[16] http://www.iiia.csic.es/en/project/eve