noisy label learning aims to train deep neural networks using a large amount
of samples with noisy labels, whose main challenge comes from how to deal with
the inaccurate supervision caused by wrong labels. Existing works either take
the label correction or sample selection paradigm to