This repository is the official implementation of SMS, and the corresponding paper is under review.
python = 3.8.17
pytorch = 2.0.0
matplotlib
numpy
...
- To run the SMS on MNIST
python /VMU/Verifiable_MU/On_MNIST/VMU_on_MNIST.py
- To run the MIB on MNIST
python /VMU/Verifiable_MU/On_MNIST/Membership_inf_via_backdoor/MIB_on_MNIST.py
- To run the SMS on CIFAR10
python /VMU/Verifiable_MU/On_CIFAR10/VMU_on_CIFAR10.py
- To run the MIB on CIFAR10
python /VMU/Verifiable_MU/On_CIFAR10/Membership_inf_via_backdoor/MIB_on_CIFAR10.py
- To run the SMS on CIFAR100
python /VMU/Verifiable_MU/On_CIFAR100/VMU_on_CIFAR100.py
- To run the SMS on CelebA
python /VMU/Verifiable_MU/On_CelebA/VMU_on_CelebA32.py
- To run the MIB on CelebA
python /VMU/Verifiable_MU/On_CelebA/MIB_method/MIB_on_CelebA.py
CelebA, MSR=0.6%
| Learning Verification | Non-Verif. | MIB | SMS |
|---|---|---|---|
| Model Acc. | 97.22% | 97.20% | 97.28% |
| Verifiability | - | 93.27% | 95.38% |
| Unambiguity | - | 57.38% | 90.77% |
| Running time (s) | 3015 | 3421 | 4225 |
| Unlearning Verification | Non-Verif. | MIB | SMS |
|---|---|---|---|
| Model Acc. | 97.15% | 97.17% | 97.26% |
| Verifiability | - | 100% | 3.41% |
| Unambiguity | - | 53.72% | 98.21% |