ZERO-VARIANCE SIMULATED ANNEALING FOR BAYESIAN SYSTEM IDENTIFICATION

Green, Peter L (2017) ZERO-VARIANCE SIMULATED ANNEALING FOR BAYESIAN SYSTEM IDENTIFICATION. In: 1st International Conference on Uncertainty Quantification in Computational Sciences and Engineering, 2017-6-15 - 2017-6-17, Rhodes Island, Greece.

Text PLG_UNCECOMP_2017.pdf - Submitted version Access to this file is embargoed until Unspecified. Download (1MB)

Abstract

Markov chain Monte Carlo (MCMC) algorithms are a set of methods which allow samples to be generated from generic probability distributions. They have been used to aid the simulation of rare events, the Bayesian system identification of systems which are nonlinear and/or are approached using a Bayesian hierarchical structure and the training of a variety of machine learning algorithms (for example). The current paper discusses the 'Zero- Variance method', which can be used to greatly reduce the sample variance of quantities that are estimated using Monte Carlo methods. The ability of this approach to increase the efficiency of gradient based MCMC methods is illustrated. Finally, a Zero-Variance version of the well-known simulated annealing algorithm is employed. The algorithm is demonstrated on the Bayesian system identification of a nonlinear dynamical system.

Item Type:	Conference or Workshop Item (Unspecified)
Depositing User:	Symplectic Admin
Date Deposited:	20 Mar 2017 10:44
Last Modified:	25 Apr 2024 23:26
DOI:	10.7712/120217.5361.17287
Related URLs:	Publisher
URI:	https://livrepository.liverpool.ac.uk/id/eprint/3006518