Understanding Self-Supervised Learning Dynamics without Contrastive Pairs

                  Yuandong Tian 1 Xinlei Chen 1 Surya Ganguli 1 2

             Abstract                quiring expensive target labels (Devlin et al., 2018). Many
                                 state-of-the-art SSL methods in computer vision employ
   While contrastive approaches of self-supervised      the principle of contrastive learning (Oord et al., 2018; Tian
   learning (SSL) learn representations by minimiz-    ...