This code repository for the paper Enhancing Robustness of Deep Neural Networks Against Adversarial Malware Samples: Principles, Framework, and AICS'2019 Challenge, Li et al., AICS 2019

Overview

Four kinds of defenses are implemented against adversarial evasion attacks produced by the AICS 2019 organiser. The descriptions of the challenge problem are put in the folder challenge. More information can be found here.

Dependencies:

• python 2.7 or python 3.6
• Codes have been tested on tensorflow-gpu==1.9.0 and Tensorflow-gpu==1.14.0
• numpy >= 1.13.3
• scikit-Learn >= 0.18.2

Usage && Files Descriptions

We are required to change the project_root in the file of conf to accommodate the current absolute path.

Run

1. Execute main.py

python main.py defender -tp

2. To reproduce the experiment results reported in the paper, we can execute main.py:

python main.py defender -d random_subspace -tp

Please follow the helper function in main.py if you'd like to execute other defenses. All learned model will be saved into the current directory under save folder which can be reset in the file of conf

Citation

If you'd like to cite us, please consider the following:

@inproceedings{li2019enhancing,
  title={Enhancing Robustness of Deep Neural Networks against Adversarial Malware Samples: Principles, Framework, and Application to AICS’2019 Challenge},
  author={Li, Deqiang and Li, Qianmu and Ye, Yanfang and Xu, Shouhuai},
  booktitle={The AAAI-19 Workshop on Artificial Intelligence for Cyber Security (AICS), 2019},
  year={2019}
}