INCORPORATING UNCERTAINTY IN NEURAL NETWORKS
Bernhard R. Kämmerer · International Journal of Pattern Recognition and Artificial Intelligence · 1992
We propose a method to incorporate the uncertainty of data in the computation process of neural networks. A measure of certainty is used on each input element in order to modulate the element's contribution to the whole input activity. The amount of certainty may result from knowledge about sensor data (e.g. detectable hardware faults or information from preprocessing steps) or may be determined in previous neurons. The method is developed and studied within the scope of the perceptron model and tested on an image processing application.