Detecting hard synapses faults in artificial neural networks

摘要

This paper presents the concepts of detecting hard faults in artificial neural network synapses using the modification of the neural network settings. The core of this work is based on weights values modification and inserting the chosen testing data when comparing the neural network output to the known valid results. The paper also discusses the problem of neural networks output saturation and provides experiments regarding an influence of the neural network settings to the problem.