attack.TrojanNN

class TrojanNN[source]

Bases: BadNet

Trojaning Attack on Neural Networks

basic structure:

  1. config args, save_path, fix random seed

  2. set the clean train data and clean test data and load pretrained model

  3. find a good trigger perturbation pattern, set the attack img transform and label transform

  4. set the backdoor attack data and backdoor test data

  5. set the device, model, criterion, optimizer, training schedule.

  6. attack or use the model to do finetune with 5% clean data

  7. save the attack result for defense

attack = TrojanNN()
attack.attack()

Note

@inproceedings{Trojannn, author = {Yingqi Liu and Shiqing Ma and Yousra Aafer and Wen-Chuan Lee and Juan Zhai and Weihang Wang and Xiangyu Zhang}, title = {Trojaning Attack on Neural Networks}, booktitle = {25th Annual Network and Distributed System Security Symposium, {NDSS} 2018, San Diego, California, USA, February 18-221, 2018}, publisher = {The Internet Society}, year = {2018},}

Parameters:

  • attack (string) – name of attack, use to match the transform and set the saving prefix of path.

  • attack_target (Int) – target class No. in all2one attack

  • attack_label_trans (str) – which type of label modification in backdoor attack

  • pratio (float) – the poison rate

  • bd_yaml_path (string) – path for yaml file provide additional default attributes

  • pretrain_model_path (string) – path for pretrained model

  • mask_path (string) – path for mask image

  • selected_layer_name (string) – name of selected layer in target model

  • selected_layer_param_name (string) – name of selected layer’s parameter in target model

  • num_neuron (int) – number of neurons to be selected in target layer

  • neuron_target_values (float) – the target value for selected neurons, you can change to list in the yaml if necessary

  • mask_update_iters (int) – number of iterations to update mask

  • resource_folder_path (string) – path for resource folder, which contains the mask image

  • **kwargs (optional) – Additional attributes.