vincentballet/LowProFool

Repository of the paper "Imperceptible Adversarial Attacks on Tabular Data" presented at NeurIPS 2019 Workshop on Robust AI in Financial Services (Robust AI in FS 2019)

LowProFool

LowProFool is an algorithm that generates imperceptible adversarial examples

This GitHub hosts the code to replicate the experiments presented in the paper:

Ballet, V., Renard, X., Aigrain, J., Laugel, T., Frossard, P., & Detyniecki, M. (2019). Imperceptible Adversarial Attacks on Tabular Data. NeurIPS 2019 Workshop on Robust AI in Financial Services: Data, Fairness, Explainability, Trustworthiness, and Privacy (Robust AI in FS 2019)

https://arxiv.org/abs/1911.03274

Adverse.py

Contains the implementation of LowProFool [1] along with an modifier version of DeepFool [2] that handles tabular datasets.

Metrics.py

Implements metrics introduced in [1]

Playground.ipynb

A demo python notebook to generate adversarial examples on the German Credit dataset and compare the results to DeepFool

References

[1] Ballet, V., Renard, X., Aigrain, J., Laugel, T., Frossard, P., & Detyniecki, M. (2019). Imperceptible Adversarial Attacks on Tabular Data. NeurIPS 2019 Workshop on Robust AI in Financial Services: Data, Fairness, Explainability, Trustworthiness, and Privacy (Robust AI in FS 2019)

[2] S. Moosavi-Dezfooli, A. Fawzi, P. Frossard: DeepFool: a simple and accurate method to fool deep neural networks. In Computer Vision and Pattern Recognition (CVPR ’16), IEEE, 2016.