EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

1606. Mean square convergence analysis of nonlinear distributed recursive estimation under heavy-tailed noise

Invited abstract in session MB-34: Optimization and learning for data science and imaging (Part II), stream Advances in large scale nonlinear optimization.

Monday, 10:30-12:00
Room: 43 (building: 303A)

Authors (first author is the speaker)

1. Manojlo Vukovic
University of Novi Sad, Faculty of Technical Sciences, Faculty of Sciences
2. Dusan Jakovetic
University of Novi Sad Faculty of Sciences
3. Dragana Bajovic
Faculty of Technical Sciences, Univ. of Novi Sad
4. Soummya Kar
Carnegie Mellon University

Abstract

Distributed recursive estimation under heavy-tailed sensing and communication noises is considered. Therein, the sensing and communication noises can be mutually correlated while independent identically distributed over iterations. A general setting is assumed where both the sensing and communication noises may have infinite variances. A consensus+innovations distributed estimator is presented, involving a general nonlinearity in both consensus and innovations update rules. We present several results on the estimator performance. It is shown that the mean squared error (MSE) of the estimation converges to zero, and moreover, an explicit MSE sublinear convergence rate is established. In addition, almost sure convergence and asymptotic normality results are derived. Analytical and numerical examples confirm that the presented method converges under the simultaneous heavy-tail communication and sensing noises, in contrast with existing estimators that break down under the same setting.

Keywords

Status: accepted


Back to the list of papers