max-andr / provably-robust-boosting Goto Github PK

View Code? Open in Web Editor NEW

49.0 6.0 11.0 7.31 MB

Provably Robust Boosted Decision Stumps and Trees against Adversarial Attacks [NeurIPS 2019]

Home Page: http://arxiv.org/abs/1906.03526

License: BSD 3-Clause "New" or "Revised" License

Dockerfile 0.30% Python 95.77% Shell 3.93%

adversarial-attacks provable-defense boosted-trees boosted-decision-stumps boosting

provably-robust-boosting's People

Contributors

Stargazers

Watchers

Forkers

intuitionmachine fagan2888 gwding mahetue ezeob002 stjordanis yanneu eduardoandrade seyuma mortal12138 nkgiants1110

provably-robust-boosting's Issues

What is the license of the code?

The paper looks super interesting and thanks for opensourcing the code!
Just wondering what is the license of and could you add it to the repo?

Problem on case when eps = 1

Hi,
I'm trying to reproduce some results reported in your paper.
When I tried the sample code in your readme, I found that when increasing the eps to a larger one, like 0.8 or 1.0(pretty large considering the data has been scaled to [0,1]), upper bound of robust test error is reported to be equal to standard error, which seems strange.

Thanks.

Recommend Projects