Sample-based Population Observers

TL;DR

This paper introduces a sample-based approach for population observers that estimates state distributions without suffering from the curse of dimensionality, bridging probabilistic frameworks with practical randomized strategies.

Contribution

It presents the first sample-based formulation of population observers, providing computationally efficient methods rooted in probabilistic and optimal transport theories.

Findings

Develops a new sample-based formulation for population observers.

Provides computational procedures avoiding curse of dimensionality.

Enhances understanding of ensemble estimation and tracking.

Abstract

In this paper, a first sample-based formulation of the recently considered population observers, or ensemble observers, which estimate the state distribution of dynamic populations from measurements of the output distribution is established. The results presented in this paper yield readily applicable computational procedures that are no longer subject to the curse of dimensionality, which all previously developed techniques employing a kernel-based approach are inherently suffering from. The novel insights that eventually pave the way for all different kinds of sample-based considerations are in fact deeply rooted in the basic probabilistic framework underlying the problem, bridging optimal mass transport problems defined on the level of distributions with actual randomized strategies operating on the level of individual points. The conceptual insights established in this paper not only yield insight into the underlying mechanisms of sample-based ensemble observers but significantly advance our understanding of estimation and tracking problems for the class of ensembles of dynamical systems in general.