Erroneous models in neural networks and their threats for formal verification.

Summary This article explains why current dependability techniques are not suitable for neural networks (NN). It also shows with an experiment that we need to justifiably trust neural networks modeling before formal verification can be used for critical applications.