Adversarial examples.

DeepFool.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "ename": "ImportError",
     "evalue": "No module named 'tflearn'",
     "output_type": "error",
     "traceback": [
      "\u001b[1;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[1;31mImportError\u001b[0m                               Traceback (most recent call last)",
      "\u001b[1;32m<ipython-input-2-358489b846ed>\u001b[0m in \u001b[0;36m<module>\u001b[1;34m()\u001b[0m\n\u001b[1;32m----> 1\u001b[1;33m \u001b[1;32mimport\u001b[0m \u001b[0mtflearn\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m      2\u001b[0m \u001b[1;32mfrom\u001b[0m \u001b[0mtflearn\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mlayers\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mcore\u001b[0m \u001b[1;32mimport\u001b[0m \u001b[0minput_data\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mdropout\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mfully_connected\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m      3\u001b[0m \u001b[1;32mfrom\u001b[0m \u001b[0mtflearn\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mlayers\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mconv\u001b[0m \u001b[1;32mimport\u001b[0m \u001b[0mconv_2d\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mmax_pool_2d\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m      4\u001b[0m \u001b[1;32mfrom\u001b[0m \u001b[0mtflearn\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mlayers\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mnormalization\u001b[0m \u001b[1;32mimport\u001b[0m \u001b[0mlocal_response_normalization\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m      5\u001b[0m \u001b[1;32mfrom\u001b[0m \u001b[0mtflearn\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mlayers\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mestimator\u001b[0m \u001b[1;32mimport\u001b[0m \u001b[0mregression\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;31mImportError\u001b[0m: No module named 'tflearn'"
     ]
    }
   ],
   "source": [
    "import tflearn\n",
    "from tflearn.layers.core import input_data, dropout, fully_connected\n",
    "from tflearn.layers.conv import conv_2d, max_pool_2d\n",
    "from tflearn.layers.normalization import local_response_normalization\n",
    "from tflearn.layers.estimator import regression\n",
    "\n",
    "# Data loading and preprocessing\n",
    "import tflearn.datasets.mnist as mnist\n",
    "X, Y, testX, testY = mnist.load_data(one_hot=True)\n",
    "X = X.reshape([-1, 28, 28, 1])\n",
    "testX = testX.reshape([-1, 28, 28, 1])\n",
    "\n",
    "# Building convolutional network\n",
    "network = input_data(shape=[None, 28, 28, 1], name='input')\n",
    "network = conv_2d(network, 32, 3, activation='relu', regularizer=\"L2\")\n",
    "network = max_pool_2d(network, 2)\n",
    "network = local_response_normalization(network)\n",
    "network = conv_2d(network, 64, 3, activation='relu', regularizer=\"L2\")\n",
    "network = max_pool_2d(network, 2)\n",
    "network = local_response_normalization(network)\n",
    "network = fully_connected(network, 128, activation='tanh')\n",
    "network = dropout(network, 0.8)\n",
    "network = fully_connected(network, 256, activation='tanh')\n",
    "network = dropout(network, 0.8)\n",
    "network = fully_connected(network, 10, activation='softmax')\n",
    "network = regression(network, optimizer='adam', learning_rate=0.01,\n",
    "                     loss='categorical_crossentropy', name='target')\n",
    "\n",
    "# Training\n",
    "model = tflearn.DNN(network, tensorboard_verbose=0)\n",
    "model.fit({'input': X}, {'target': Y}, n_epoch=20,\n",
    "           validation_set=({'input': testX}, {'target': testY}),\n",
    "           snapshot_step=100, show_metric=True, run_id='convnet_mnist')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.5.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}

Notes.md

Related links

• Analysis of classifiers’ robustness to adversarial perturbations
• DeepFool

Questions

• What if I do argmax(r) = ||r||2 s.t. f(x+r)f(x) >= 0