Normalized Loss Functions for Deep Learning with Noisy Labels