Awesome

This is a Python implementation of "Towards Unbounded Machine Unlearning"

The main experiments

• Use small_scale_unlearning.ipynb for:
  • small-scale experiemnts
  • large forget-set size experiemnts
• Use large_scale_unlearning.ipynb for:
  • large-scale experiments
• Use small_scale_ictest.ipynb for:
  • Interclass Confusion Metric experiemnts from pdf
• Use large_scale_ictest.ipynb for:
  • Interclass Confusion Metric experiemnts from pdf
• Use MIA_experiments.ipynb for:
  • Membership Inference Attack based on the model's loss values

Models choices

• For small-scale experiments:
  • allcnn --filters = 1.0
  • resnet --filters = 0.4
• For large-scale experiments:
  • allcnn --filters = 1.0
  • resnet --filters = 1.0

Datasets choices

• For small-scale datasets:
  • small_cifar5
  • small_lacuna5
• For large-scale datasets:
  • cifar10
  • lacuna10

References

We have used the code from the following two repositories:

(Selective Forgetting)[https://github.com/AdityaGolatkar/SelectiveForgetting.git]

(RepDistiller)[https://github.com/HobbitLong/RepDistiller.git]