AAAI24 paper

This is the official repository for "Game-Theoretic Unlearnable Example Generator"). This repository contains an implementation of GUE and evaluation on poisoned datasets.

Requirements:

• Python 3.11.3
• PyTorch 2.0.1
• Torchvision 2.0.0

Running experiments:

We give an example of training GUE generator on CIFAR-10 dataset:

python gue.py --dataset cifar10 --epochs 50 --eta 1.5 --lr_atk 0.1  --lr_cls 0.01 --tensorboard_path save/path

Evaluate the unlearnable examples generated by GUE generator：

python evaluation.py --dataset cifar10 --arch ResNet18 --out_dir ./results --path ./your/path/atkmodel --exp_name gue  --train_loss ST  --lr 0.01

We also provide pretrained GUE generator. To test it, use:

python evaluation.py --dataset cifar10 --arch ResNet18 --out_dir ./results --path ./gue_cifar10.pth --exp_name gue  --train_loss ST  --lr 0.01