Figure 4.
The flowchart begins with Self-Supervised Adversarial Training under Motivation. It branches under Models to Contrastive Learning and Adversarial Training. These lead under Problems to Inconsistent Class Definition and Inconsistent Distribution Bias. The problems connect under Implementation to Classification in Instance Discrimination and Separate Views Anchors slash Classes. Both implementations lead to the Extended Solution stage. This contains Two-View Contrastive Learning and, under Target Solution, Two-View Adversarial Training.Illustration of how self-supervised solutions to two-view contrastive learning and two-view adversarial training are developed