tttjjjwww/hashingbasednn

Hashing-based DNNs defend against adversarial evasions in malware detection

---

Hashing-based Deep Neural Network

This project provides the coresponding code to the paper "Resisting Adversarial Examples using Hashing-based Deep Neural Networks in Malware Detection"

Dependencies:

• Python 2.7
• Numpy 1.11.3
• Matplotlib 2.0.0
• Tensorflow 1.3 or 1.4
• Scikit-Learn 1.0.0
• Jupyter notebook

Usage && Files Descriptions

In root direcotory, we do joyful examples on synthesis dataset:

• generate adversarial examples:
  gen-adv-smps.ipynb
• construct DNN graphs
  graphs.py
• Multi-index hashing based DNNs
  InH.ipynb
• Local forest hashing based DNNs
  LFH.ipynb
• Joint index hashing and Denoising auto-encoder
  JID.ipynb
• Joint locah forest hashing and Denoising auto-encoder
  JFD.ipynb
• other files
  utils.py learning_hashing_by_RF.py

We recommend to run gen-adv-smps.ipynb first to obain adversarial examples, and then perform any one of InH.ipynb, LFH.ipynb, JID.ipynb and JFD.ipynb

In drebin directory:

There are source codes for a series expriments on drebin. We need apply the DREBIN dataset at here: drebin.

In pdfrate directory:

We provide the robust support vector machine method.

FAQ

We conduct the experiments on the CPU server which 64 cores CPU (2.4GHz) and shared 150G RAM.