Wisdom ofthe Ensemble: Improving Consistency of Deep Learning Models