ResearchBib Share Your Research, Maximize Your Social Impacts
Sign for Notice Everyday Sign up >> Login

ONE INCREMENTAL ADAPTATION STRATEGY OF THE DECISION-MAKING FORMAL NEURON

Journal: International Journal of Engineering Sciences & Research Technology (IJESRT) (Vol.4, No. 1)

Publication Date:

Authors : ; ;

Page : 333-346

Keywords : vote-weight; Mahalanobis distance; entropy sensitivity criterion; Widrow-Hoff algorithm; RobbinsMonro algorithm.;

Source : Downloadexternal Find it from : Google Scholarexternal

Abstract

The general objective of formal neuron adaptation is to give the reliable inputs more influence in determining the output than the unreliable inputs have. Adaptation is introduced into the threshold decision element by a circuit that performs two operations: estimates the error probability of each input and uses the estimate to change each voteweight. A cyclic error-counting adaptation procedure is the one in which vote-weights are changed periodically (cyclically), based on data collected during computations in a period (cycle). If appropriate incremental adjustments are made to the estimate of inputs' each error after each computation, then we deal with continuous adaptation. There are two methods for detecting errors: comparison with the output decision (closed-loop adaptation) and comparison with an externally supplied correct answer (open-loop adaptation). In this paper, the input weighted sum is compared with the desired value of this sum, which is multiplied by the correct binary answer. Our results for incremental changes of the input weights of the adaptive formal neuron are based on the Widrow-Hoff algorithm and stochastic approximation method.

Last modified: 2015-02-09 22:01:08