An Inclusive Theoretical Framework of Robust Supervised Contrastive Loss against Label Noise