Figure 9:

Cora-ML. Clustering of training nodes’ embeddings generated by private models with different privacy guarantees $ϵ$ (fixed $=2\times {10}^{-3}$) and training methods.