nico-adamo/adversarialensemble

Measuring the transfer of adversarial examples over ensemble networks of the same architecture (VGG-16 and LeNet-5)

AdversarialEnsemble

Measuring the transfer of adversarial examples over ensemble networks of the same architecture (VGG-16 and LeNet-5) Testing the theory of whether adversarial examples take on arbitrary labels under the same models trained w/ different initializations.

models contains an abstract base class implementation of an ensemble network, as well as the specific instantiations for VGG-16 on Cifar and LeNet on MNIST. models/nn contains an implementation of LeNet-5 and VGG-16 - shouldn't be too hard to add other models.

Re-implementation from Mathematica, found here: https://github.com/Nico-Adamo/WSC2019-Colorful-Fraud