Code for the paper "Backdoor Attacks Against Vertical Split Learning". Now we are refining the journal editions and will open source soon.
usage: label_inference_and_backdoor_attack.py [-h] [--dataset DATASET]
[--target_label TARGET_LABEL]
[--target_label_candidates_number TARGET_LABEL_CANDIDATES_NUMBER]
[--test_parameter TEST_PARAMETER]
[--test_noise TEST_NOISE]
[--test_trigger_fabrication TEST_TRIGGER_FABRICATION]
[--test_sample_n TEST_SAMPLE_N]
[--test_server_layer TEST_SERVER_LAYER]
[--test_lr TEST_LR]
[--noise_range NOISE_RANGE]
[--noise_p NOISE_P]
[--mask_size MASK_SIZE]
[--poisoning_rate POISONING_RATE]
[--wave_multiple WAVE_MULTIPLE]
[--backdoor_epochs BACKDOOR_EPOCHS]
[--upload_method UPLOAD_METHOD]
You can run the code as follow:
python -u label_inference_and_backdoor_attack.py --dataset MNIST --backdoor_epochs 10
python -u label_inference_and_backdoor_attack.py --dataset imagenette --backdoor_epochs 100
python -u label_inference_and_backdoor_attack.py --dataset Cifar10
python -u label_inference_and_backdoor_attack.py --dataset cinic10
python -u label_inference_and_backdoor_attack.py --dataset bank
python -u label_inference_and_backdoor_attack.py --dataset givemesomecredit
usage: data_split_backdoor.py [-h] [--dataset DATASET]
[--target_label TARGET_LABEL]
[--target_label_candidates_number TARGET_LABEL_CANDIDATES_NUMBER]
[--noise_range NOISE_RANGE] [--noise_p NOISE_P]
[--mask_size MASK_SIZE]
[--poisoning_rate POISONING_RATE]
[--wave_multiple WAVE_MULTIPLE]
[--backdoor_epochs BACKDOOR_EPOCHS]
[--n_workers N_WORKERS]
[--id_attacker ID_ATTACKER]
You can run the code as follow:
python -u data_split_backdoor.py --dataset Cifar10 --n_workers 4 --id_attacker 1