Modeling, estimation and identification of stochastic systems with latent variables
Year: 2013
Type: article
Abstract: The main topic of this thesis is the analysis of static and dynamic models in which some variables, although directly influencing the behavior of certain observables, are not accessible to measurements. These models find applications in many branches of science and engineering, such as control systems, communications, natural and biological sciences and econometrics. It is well-known that models with unaccessible - or latent - variables, usually suffer from a lack of uniqueness of representation. In other words, there are in general many models of the same type describing a given set of observables say, the measurable input-output variables. This is well-known and has been well-studied for a special class of linear models, called state-space models. In this thesis we shall focus on two particular classes of stochastic systems with latent variables: the generalized factor analysis models and errors-in-variables models. For these classes of models there are still some unresolved issues related to non-uniqueness of the representation and clarifying these issues is of paramount importance for their identification. Since mathematical models usually need to be estimated from experimental data, solving the non-uniqueness problem is essential for their use in statistical inference (system identification) from measured data.
Author: Giulio Bottegal
Language: English
Subfield: Control and Systems Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W8526304