Wisdom of the Ensemble: Improving Consistency of Deep Learning Models