A NEURAL NETWORK SYSTEM FOR THE SELF-LEARNING RECOGNITION OF HAND-DRAWING LOGIC DIAGRAM
Ueliang Shanghai · Computer Applications and Software · 1992
We describe a method for solving nonlinear supervised learning tasks bymaking forward broadcast discretized and globally training weight arrays in themultilayer neural network by the affirmation-negation way with guide of the stocha-stic variable of output mean squared error.The discretized forward broadcast decrea-ses in number of kinds of the hold-value on neurons.The algorithm of self-learningis gobally simplified by the affirmation-negation.The method dispenses with the morecomplicated back-propagation computation,and yet it dosen't lose its learning speed.This method is called as DSGNN algorithm.DSGNN may be more suitable to parallelimplementation and biological perspective simulation.DSGNN has been successfullyapplied to the system for self-learning recognition of hand-drawing logic diagram.The rate of recognition accuracy is above 95%.