Towards Robust Learning with Different Label Noise Distributions