Título: Stochastic Parameter Estimation Neural Nets Supervised Learning Approach
Autores: Rios Neto, Atair
Resumo: In this short paper the problem of supervised learning of a neural net is treated as one of stochastic parameter estimation. The identification of the neural net parameters in its supervised training is as usually formulated as an optimization problem. A linear pertubation scheme, reducing the problem to one of stochastic parameter estimation in each iteration is then proposed. The structure of a Kalman filtering in the resulting algorithm allows the use of all the existing results in state and parameter estimation to guarantee a good performance to the proposed procedure.
Código DOI: 10.21528/CBRN1994-010
Artigo em PDF: CBRN1994-paper10.pdf
Arquivo BibTex: CBRN1994-paper10.bib