Title :
Error tolerant multilayer perceptrons with incompletely specified inputs
Author_Institution :
IBM Corp., San Jose, CA, USA
Abstract :
Summary form only given, as follows. The backpropagation learning algorithm of the multilayer perceptron has produced solutions for weight matrices of the linkages with the disadvantage that it does not reveal the structure of the multilayer perceptrons for performing further analysis and enhancement. A controlled learning environment of multilayer perceptrons has been developed and reported by the authors. This study further enhances the learning environment by allowing input errors as well as incompletely specified inputs. Under the input environment with error, an algorithm to modify the threshold and weightage functions is developed which will lead to the right output solution. Under the environment with incompletely specified input, an algorithm to determine a set of new threshold functions is also developed so that incompletely specified inputs can be tolerated to produce matched output solutions.<>
Keywords :
hierarchical systems; learning systems; neural nets; backpropagation learning algorithm; controlled learning environment; error-tolerant perceptrons; multilayer perceptrons; threshold functions; weight matrices; weightage functions; Hierarchical systems; Learning systems; Neural networks;
Conference_Titel :
Neural Networks, 1989. IJCNN., International Joint Conference on
Conference_Location :
Washington, DC, USA
DOI :
10.1109/IJCNN.1989.118353