Assessing the Reliability of Deep Learning Applications