Adversarial examples

Requirements:

• Python 3
• numpy
• tensorflow (tested with versions 1.2 and 1.4)
• scipy (tested with version 0.19.1)

Attacks

• JSMA (notebook)(code).

This part we cite the work of Papernot et al.. Default model in the source code is a deep neural network defined in above respository. This part relies on cleverhans's other files, you my need to install the whole respository for running this code.

• FGSM (notebook)(code).

We also cite this work from cleverhans.This tutorial covers how to train a MNIST/CIFAR model using TensorFlow, craft adversarial examples using the fast gradient sign method, and make the model more robust to adversarial examples using adversarial training.

• CW attack (notebook)(code).

"Towards Evaluating the Robustness of Neural Networks" by Nicholas Carlini and David Wagner, at IEEE Symposium on Security & Privacy, 2017.

CW attack consists of L0 attack,L2 attack and Li attack. In our work, we only test L2 attack.This tutorial covers how to train a MNIST model using TensorFlow, craft adversarial examples using CW attack, and prove that defensive distillation is not robust to adversarial examples.More details in Nicholas Carlini et al..

• Fast feature fool (notebook)(code).

Test fast feature fool algorithm with MNIST dataset has not been finished yet, there's the source code of Mopuri et al..

• Box-constrained attacks (notebook)(code).

NIPS 2017 adversarial attacks/defenses competition:

• pgd_attack: Uses projected SGD (Stochastic Grandient Descent) as optimizer
• step_pgd_attcK: Uses a mix of FGSM (Fast Gradient Sign Attack) and SGD. We found this to converge faster if there is a limit of only a few iterations (e.g. 10-15)

For a more comprehensive example, please check the provided luizgh/adversarial_examples

Physical-World Attacks

Robust Physical-World Attacks on Deep Learning Models

Article

Traffic Sign Classifier

US LISA Dataset

The code in this repository is helpful to Convert the LISA Traffic Sign dataset into Tensorflow tfrecords.

1. Download LISA Dataset here : http://cvrr.ucsd.edu/LISA/lisa-traffic-sign-dataset.html

2. But only uses 17 classes in this project, as shown in categories.txt

German gtsrb Dataset

Download the dataset. You can download the pickled dataset in which we've already resized the images to 32x32,

Here

Training

python3 lisa-classifier.py

pyrhon3 gtsrb-classifier.py

Test

demo.ipynb

Reference

• Traffic Sign Classification using Convolutional Neural Networks