n-ML: Mitigating Adversarial Examples via Ensembles of Topologically Manipulated Classifiers

robust-ml / robust-ml.github.io

A community-run reference for state-of-the-art adversarial example defenses.

https://www.robust-ml.org/

Creative Commons Attribution Share Alike 4.0 International

49 stars 7 forks source link

n-ML: Mitigating Adversarial Examples via Ensembles of Topologically Manipulated Classifiers #16

Open Apromixately opened 4 years ago

Apromixately commented 4 years ago

Name: n-ML: Mitigating Adversarial Examples via Ensembles of Topologically Manipulated Classifiers Authors: Mahmood Sharif, Lujo Bauer, Michael K. Reiter

Paper: https://arxiv.org/pdf/1912.09059.pdf

Code: -

Venue: -

Does the code implement the robust-ml API and include pre-trained models: -

Dataset: MNIST, CIFAR10, GTSRB

Threat model: white box, gray box, black box

Natural accuracy: e.g. CIFAR10 / black box / L_inf <= 8/255: 94.50 %

Claims: e.g. CIFAR10 / black box / L_inf <= 8/255: 100.00 %

anishathalye commented 4 years ago

Thank you for the submission! (Are you one of the authors? I can't guess from your GitHub profile.)

As per our policy, in order to be listed, defenses must have code publicly available (with pre-trained models, and implementing the robust-ml API). Do you know if there's code available for this defense?

Apromixately commented 4 years ago

I am not one of the authors. I've sent an email to Mahmood last week but haven't gotten a reply, yet.

Apromixately commented 4 years ago

Oh, in case you haven't had a look: they say in the paper that they will publish the code when publishing the paper. So it might just not have happened yet or be postponed until it is accepted at a conference.