Graph structure and recursive estimation of noisy linear relations

Abstract : This paper examines estimation problems specified by noisy linear relations describing either dynamical models or measurements. Each such problem has a graph structure, which can be exploited to derive recursive estimation algorithms only when the graph is acyclic, i.e. when it is obtained by combining disjoint trees. Aggregation techniques appropriate for reducing an arbitrary graph to an acyclic one are presented. The recursive maximum likelihood estimation procedures that we present are based on two elementary operations, called reduction and extraction, which are used to compress successive observations and discard unneeded variables. These elementary operations are used to derive filtering and smoothing formulas applicable to both linear and arbitrary trees, which are in turn applicable to estimation problems in settings ranging from 1-D descriptor systems to 2-D difference equations to multiscal models of random fields. These algorithms can beviewed as direct generalizations to a far richer setting of Kalman filtering and both two-filter and Rauch-Tung-Striebel smoothing for standard causal state space models.
Type de document :
[Research Report] RR-1912, INRIA. 1993
Liste complète des métadonnées
Contributeur : Rapport de Recherche Inria <>
Soumis le : mercredi 24 mai 2006 - 16:15:08
Dernière modification le : vendredi 25 mai 2018 - 12:02:05
Document(s) archivé(s) le : mardi 12 avril 2011 - 19:08:07



  • HAL Id : inria-00074761, version 1



Ramine Nikoukhah, D. Taylor, Bernard C. Levy, A.S. Willsky. Graph structure and recursive estimation of noisy linear relations. [Research Report] RR-1912, INRIA. 1993. 〈inria-00074761〉



Consultations de la notice


Téléchargements de fichiers