As ML is being used in more places, including mission critical systems, it is important to pay attention to corner cases that could fail. In this paper the authors study memory faults and present a solution for this problem when neural networks are being used. The solutions presented are straight forward however the novelty comes from introducing the problem to the ML communitty and by presenting solutions that take into account the specifics of the ML task. Therefore, this work can ignite an interesting research direction.