Room 206 (2nd floor, badged access)
30 mai 2024 - 14h00
A Hybrid-Systems Framework for Distributed Gradient-Based Estimation
par Mohamed Maghenem de CNRS, GIPSA-lab, Grenoble
Abstract: We address the classical problem of estimating the constant unknown parameters of a given linear input/output relationship.
Motivated by large-scale estimation problems and concurrent learning, the proposed method uses a network of gradient-descent-based estimators, each of which explores only a subset of (local) input-output data.
A key feature of the method is that the input-output signals are allowed to be hybrid, so they may evolve continuously ( i.e., they may flow), and they may instantly change at isolated time instances (i.e., they may jump).
The estimators exchange their estimates according different protocols characterised by a weakly-connected directed graph.
As a result, a condition of persistency of excitation, in a hybrid and distributed form, is shown to ensure exponential convergence of the estimation errors.
