Fragility in Artificial Intelligence Systems Based on Artificial Neural Networks

What is it about?

Artificial neural networks (ANNs) often exhibit fragility: small changes in input data cause them to behave in undesirable and unpredictable ways, such as mis-identifying images or mis-classifying situations. Based on what is known about ANNs and about natural neural networks, i.e., animal nervous systems, we explain ANN fragility and offer possible remedies.

Featured Image