Becoming More Robust to Label Noise with Classifier Diversity