Adversarial Attack Using Genetic Algorithm

This is a repo that accompanies my article about generating adverasarial examples. It consists of:

• GeneticSolver.py - general framework for genetic approach rewritten for MNIST dataset
• ImageGeneticSolver.py - Same as GeneticSolver, but saves the best candidate so far in verbose mode
• ga_adv.ipynb - main notebook with the results described in article
• ga_adv_robust.ipynb - bonus - exploration of the idea of multi-task learning. If the attacker knows that you consider the example to be fake, he can still use this information during optimization
• mnist.pnz - MNIST dataset for training