Supplementary Materials for " Private Set Generation with Discriminative Information "

Our privacy computation is based on the notion of Rényi-DP, which we recall as follows. Lastly, we use the following theorem to convert ( α,ε) -RDP to (ε, δ) -DP . The total dataset size is 60K for the training set and 10K for the testing set, respectively. All our models and methods are implemented in PyTorch. Based on Google's TensorFlow privacy under version The experiments presented in Section 5.2 of the main paper correspond to the class-incremental learning setting [ And the task protocol is sequentially learning to classify a given sample into all the classes seen so far.