ML Classifier with Adversarial Images

---

Create adversarial images that can be used to fool to misclassify MNIST data in a deep convolutional neural network.

Example TensorFlow

https://github.com/lucylow/ML_adversarial_images/blob/master/MNIST_CNN_working.ipynb

Prerequisites

• Python 3.3 +
• Tensorflow 1.1.0
• matplotlib 2.1.0
• numpy

Related

• [1] TensorFlow's 'Deep MNIST for Experts' guide
• [2] Goodfellow's 'Explaining and Harnessing Adversarial Examples'
• [3] Andrej Karpathy's 'Breaking Linear Classifiers on ImageNet'