Optimal Estimation by using Neural Networks

Authors:	Stepanov Oleg, CSRI Elektropribor, Russian Federation Amosov Oleg, Komsomolsk-on-Amur State Technical University, Russian Federation
Topic:	1.4 Stochastic Systems
Session:	Linear and Nonlinear Filtering in Stochastic Systems
Keywords:	minimum variance estimation, non-Gaussian a posteriori probability density function, neural network, comparison

Abstract

The relation between the traditional (minimum variance) algorithms for estimation of random vectors and the algorithms based on the use of neural networks has been investigated. It is shown that the Bayesian and neural network algorithms provide estimates with similar properties. The results derived are discussed. The examples (in particular problem with a non-Gaussian a posterior probability function) are considered.