Apply Membership Inference Attack and Differential Privacy on ML Binary Classifier

Small dataset: Start with 10K and then scale it up.
Step 1. Train a normal deep learning classifier with a duplication based dataset (non-image dataset)
Step 2. Run Membership Inference Attack (MIA) on this classifier

Step 3. Add Differential Privacy (DP) to train a new model with or without duplicate datasets.
Apply step 2 on it.

In total 4 classifier needs to be trained.
Classifiers to train:
1. Classifier trained on data WITHOUT non-reoccurring information within samples and WITHOUT DP (+MIA)
2. Classifier trained on data WITH non-reoccurring information within samples and WITHOUT DP (+MIA)
3. Classifier trained on data WITHOUT non-reoccurring information within samples and WITH DP (+MIA)
4. Classifier trained on data WITH non-reoccurring information within samples and WITH DP (+MIA)

Other details will be provided later.