A robustness framework for baseline of standard and adversarial machine learning research projects
The robustness framework is based on top-tier machine learning libraries based on Pytorch. Additionally, it allows for embedding different model architectures and training processes in addition to fixing all research issues. This framework integrates the following libraries:
- Pytorch-Lightning
- Hydra
- torchattacks
The following architectures are covered in this framework and other networks will be added as needed:
- MKToyNet
- LeNet
- DenseNet
- ResNet
- WideResNet
The logging system of this framework is also customizable and resolves all your ideas. As a result, we can take advantage of the efficiency of the following libraries:
- TorchMetrics
- Loggings
- Neptune
- Comet
- MLFlow
- ...
To install this interesting framework for standard and adversarial machine learning, follow the steps below, and don't waste your time developing an efficient baseline. For installing robustness framework, we have two approaches:
git clone https://github.com/khalooei/robustness-framework.git
pip install -r requirements.txt
pip install robustness-framework
You can just follow the main.py file as a main anchor of this framework. You can define your own configurations in configs directory as we defined training_mnist.yaml and training_cifar10.yaml configuration.
You can run this framework for running on CIFAR10 dataset as below:
python main.py +configs=training_cifar10
If you want to change the default training_cifar10 configurations, you can pass them as below. Below is an example of adversarial training using a PGD attack.
python main.py +config=training_cifar10 training_params.type="AT" adversarial_training_params.name="PGD"
[TOBE COMPLETED]
Thanks to the people behind Pytorch, Lightning, torchattacks, hydra, and MLOps libraries whose work inspired this repository. Furthermore, I would like to thank my supervisors Prof. Mohammad Mehdi Homayounpour and Dr. Maryam Amirmazlagnani for their efforts and guidance.
If you use this package, please cite the following BibTex (SemanticScholar, GoogleScholar):
@article{mkhalooei2023robustnessframework,
title={Robustness Framework: A pytorch repository for baselines of standard and adversarial machine learning research projects},
author={Mohammad Khalooei and Mohammad Mehdi Homayounpour and Maryam Amirmazlaghani},
url = {https://github.com/khalooei/robustness-framework},
year={2023}
}