mxbi / backdoor

Part II Project: Backdoors in Neural Networks

Backdoors in Neural Networks

Code for my undergraduate 3rd year project (as well as my Architectural Backdoors in Neural Networks paper: https://arxiv.org/abs/2206.07840).

The backdoor library can be found in the backdoor/ folder.

Installation is easy, just pip install ..

• For architectural backdoors, see backdoors/models.py.
• For a reimplementation of Handcrafted Backdoors in Deep Neural Networks (Hong et al.) see models/handcrafted.py.

Everything in this repo is released under MIT license, unless specified otherwise.

Setting up for development

conda create -f environment.yml # Anaconda environment used for development
pip install -e . # Editable install of the backdoor library

Running tests

pytest --cov=backdoor tests/