Randomized averaging algorithms, when can errors and unreliabilities just be ignored ?

Présenté par : Julien M. Hendrickx,
ICTEAM – Institute of Information and Communication Technologies, Electronics and Applied Mathematics

EPL – Ecole Polytechnique de Louvain
UCL – Université catholique de Louvain

5 novembre 2013 à 11h00 salle A01 à TELECOM SudParis à Evry

Abstract :

We consider randomized discrete-time consensus systems that preserve the average « on average », and provide a new upper bound on the mean square deviation of the consensus value from the initial average.

We show that a certain asymptotic accuracy can be guaranteed when there are few simultaneous interactions or when the simultaneous interactions are sufficiently uncorrelated.

Our results are easily applicable to many classes of systems, and we particularize them to various algorithms having been proposed in the literature, obtaining bounds that match or outperform the previously available ones.

Contact : Jérémie Jakubowicz