Exploring Model Dynamics for Accumulative Poisoning Discovery

This repo contains the sample code of our proposed Memorization Discrepancy and the corresponding Discrepancy-awared Sample Correction (DSC). The code is developed based on Accuulative Attack.

Environment preliminaries

This project is tested under the following environment settings:

• GPU: Geforce 3090 or Tesla V100
• Cuda: 11.4
• Python: 3.6
• PyTorch: >= 1.9.1
• Torchvision: >= 0.10.1

Running commands

Burn-in phase

python train_cifar.py

Accumulative poisoning attacks in online learning cases

Below is the original running commands for accumulative phase + poisoned trigger(controlled by --use_advtrigger) + online poisoned trigger (controlled by --use_online_advtrigger):

python online_accu_train_adv_relate.py \
                  --batch_size 100 --epoch 100 --test_batch_size 500 --log_name log_test_online.txt\
                  --resume checkpoints_base_bn --use_bn --model_name epoch40.pth \
                  --mode 'eval' --onlinemode 'train' --lr 1e-1 --momentum 0.9 \
                  --beta 1. --only_reg --threshold 0.18 --use_advtrigger

ST for Accumulative poisoning attacks

CUDA_VISIBLE_DEVICES='0' python online_accu_train_adv_relate.py \
                  --batch_size 100 --epoch 100 --test_batch_size 500 --log_name log_test_online_adv.txt\
                  --resume checkpoints_base_bn --use_bn --model_name epoch40.pth \
                  --mode 'eval' --onlinemode 'train' --lr 1e-1 --momentum 0.9 \
                  --beta 1. --only_reg --threshold 0.18 --use_advtrigger --med="ST"

AT for Accumulative poisoning attacks

CUDA_VISIBLE_DEVICES='0' python online_accu_train_adv_relate.py \
                  --batch_size 100 --epoch 100 --test_batch_size 500 --log_name log_test_online_adv.txt\
                  --resume checkpoints_base_bn --use_bn --model_name epoch40.pth \
                  --mode 'eval' --onlinemode 'train' --lr 1e-1 --momentum 0.9 \
                  --beta 1. --only_reg --threshold 0.18 --use_advtrigger --med="AT"

DSC for Accumulative poisoning attacks

CUDA_VISIBLE_DEVICES='0' python online_accu_train_adv_relate.py \
                  --batch_size 100 --epoch 100 --test_batch_size 500 --log_name log_test_online_adv.txt\
                  --resume checkpoints_base_bn --use_bn --model_name epoch40.pth \
                  --mode 'eval' --onlinemode 'train' --lr 1e-1 --momentum 0.9 \
                  --beta 1. --only_reg --threshold 0.18 --use_advtrigger --med="OURS"

Reference Code

• Accumulative Attack: https://github.com/ShawnXYang/AccumulativeAttack

• TRADES: https://github.com/yaodongyu/TRADES

---

If you find our paper and repo useful, please cite our paper:

@inproceedings{zhu2023unleashing,
title       ={Exploring Model Dynamics for Accumulative Poisoning Discovery},
author      ={Jianing Zhu and Xiawei Guo and Jiangchao Yao and Chao Du and Li He and Shuai Yuan and Tongliang Liu and Liang Wang and Bo Han},
booktitle   ={International Conference on Machine Learning},
year        ={2023}
}