Optimality of transformations for parameter estimation
Steven J. Apollo, MICHAEL T. MANRY, L.S. Allen, W.D. Lyle · 2003
Neural-network-based minimum-mean-square (MMS), maximum-a-posteriori (MAP), and maximum-likelihood (ML) parameter estimation are considered. The multilayer perceptron (MLP) is shown to approximate the minimum mean square estimator. Linear transforms are used to compress data for the purpose of efficient parameter estimation. Raw and transform domain lower bounds are developed and used as an optimality criterion in a procedure defined to compare various transformations.>