mnist_attack_demo.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# MNIST Attack"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.optim as optim\n",
    "import torchvision.transforms as transforms\n",
    "\n",
    "from torchsummary import summary\n",
    "from torch.autograd import Variable\n",
    "from torchvision.datasets import MNIST\n",
    "\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from copy import deepcopy\n",
    "from IPython.display import clear_output\n",
    "\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Building the Classifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "class ConvNet(nn.Module):\n",
    "    def __init__(self):\n",
    "        super(ConvNet, self).__init__()\n",
    "        self.features = nn.Sequential(\n",
    "            nn.Conv2d(1,  5, 5),\n",
    "            nn.BatchNorm2d(5),\n",
    "            nn.ReLU(),\n",
    "            nn.MaxPool2d(2),\n",
    "            nn.Conv2d(5, 15, 3),\n",
    "            nn.BatchNorm2d(15),\n",
    "            nn.ReLU(),\n",
    "            nn.MaxPool2d(2),\n",
    "        )\n",
    "        self.classifier = nn.Sequential(\n",
    "            nn.Linear(375, 64),\n",
    "            nn.BatchNorm1d(64),\n",
    "            nn.ReLU(),\n",
    "            nn.Linear(64, 10),\n",
    "            nn.BatchNorm1d(10),\n",
    "            nn.LogSoftmax()\n",
    "        )\n",
    "    \n",
    "    def forward(self, x):\n",
    "        out = self.features(x)\n",
    "        out = out.view(-1, 375)\n",
    "        out = self.classifier(out)\n",
    "        return out\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "trans = transforms.Compose([\n",
    "    transforms.ToTensor(), \n",
    "    transforms.Normalize((0.5,), (1.0,))\n",
    "])\n",
    "\n",
    "train_set = MNIST(root='./data', train=True,  transform=trans, download=True)\n",
    "test_set  = MNIST(root='./data', train=False, transform=trans, download=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch_size = 16\n",
    "\n",
    "train_loader = torch.utils.data.DataLoader(\n",
    "    dataset=train_set,\n",
    "    batch_size=batch_size,\n",
    "    shuffle=True\n",
    ")\n",
    "\n",
    "test_loader = torch.utils.data.DataLoader(\n",
    "    dataset=test_set,\n",
    "    batch_size=batch_size,\n",
    "    shuffle=False\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "ConvNet(\n",
       "  (features): Sequential(\n",
       "    (0): Conv2d(1, 5, kernel_size=(5, 5), stride=(1, 1))\n",
       "    (1): BatchNorm2d(5, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    (2): ReLU()\n",
       "    (3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
       "    (4): Conv2d(5, 15, kernel_size=(3, 3), stride=(1, 1))\n",
       "    (5): BatchNorm2d(15, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    (6): ReLU()\n",
       "    (7): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
       "  )\n",
       "  (classifier): Sequential(\n",
       "    (0): Linear(in_features=375, out_features=64, bias=True)\n",
       "    (1): BatchNorm1d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    (2): ReLU()\n",
       "    (3): Linear(in_features=64, out_features=10, bias=True)\n",
       "    (4): BatchNorm1d(10, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    (5): LogSoftmax()\n",
       "  )\n",
       ")"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model = ConvNet()\n",
    "model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "----------------------------------------------------------------\n",
      "        Layer (type)               Output Shape         Param #\n",
      "================================================================\n",
      "            Conv2d-1            [-1, 5, 24, 24]             130\n",
      "       BatchNorm2d-2            [-1, 5, 24, 24]              10\n",
      "              ReLU-3            [-1, 5, 24, 24]               0\n",
      "         MaxPool2d-4            [-1, 5, 12, 12]               0\n",
      "            Conv2d-5           [-1, 15, 10, 10]             690\n",
      "       BatchNorm2d-6           [-1, 15, 10, 10]              30\n",
      "              ReLU-7           [-1, 15, 10, 10]               0\n",
      "         MaxPool2d-8             [-1, 15, 5, 5]               0\n",
      "            Linear-9                   [-1, 64]          24,064\n",
      "      BatchNorm1d-10                   [-1, 64]             128\n",
      "             ReLU-11                   [-1, 64]               0\n",
      "           Linear-12                   [-1, 10]             650\n",
      "      BatchNorm1d-13                   [-1, 10]              20\n",
      "       LogSoftmax-14                   [-1, 10]               0\n",
      "================================================================\n",
      "Total params: 25,722\n",
      "Trainable params: 25,722\n",
      "Non-trainable params: 0\n",
      "----------------------------------------------------------------\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/site-packages/torch/nn/modules/container.py:91: UserWarning: Implicit dimension choice for log_softmax has been deprecated. Change the call to include dim=X as an argument.\n",
      "  input = module(input)\n"
     ]
    }
   ],
   "source": [
    "summary(model, (1, 28, 28))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "optimizer     = optim.Adadelta(model.parameters())\n",
    "loss_function = nn.NLLLoss()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/site-packages/torch/nn/modules/container.py:91: UserWarning: Implicit dimension choice for log_softmax has been deprecated. Change the call to include dim=X as an argument.\n",
      "  input = module(input)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.2968102327605079\t0.10664021118618558\n",
      "0.10602050387648224\t0.06629995576553042\n",
      "0.08051582716461271\t0.05071797690102828\n",
      "0.06920259381188071\t0.043045568852242956\n",
      "0.061306453929538714\t0.03461668851997242\n"
     ]
    }
   ],
   "source": [
    "for i in range(5):\n",
    "    #training loop\n",
    "    mean_loss = 0.\n",
    "    for j, (x, y) in enumerate(train_loader):\n",
    "        optimizer.zero_grad()\n",
    "        pred = model(x)\n",
    "        loss = loss_function(pred, y)\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "        mean_loss = mean_loss + (loss.item() - mean_loss) / float(j + 1)\n",
    "    print(mean_loss, end='\\t')\n",
    "    #testing loop\n",
    "    mean_loss = 0.\n",
    "    for j, (x, y) in enumerate(test_loader):\n",
    "        optimizer.zero_grad()\n",
    "        pred = model(x)\n",
    "        loss = loss_function(pred, y)\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "        mean_loss = mean_loss + (loss.item() - mean_loss) / float(j + 1)\n",
    "    print(mean_loss)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Building Attack\n",
    "\n",
    "https://arxiv.org/pdf/1312.6199.pdf"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "sample = x[1].detach().numpy()[0]\n",
    "\n",
    "plt.title(\"sample\")\n",
    "plt.axis('off')\n",
    "plt.imshow(sample, cmap='gray')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor(1.00000e-03 *\n",
      "       4.1220)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/site-packages/torch/nn/modules/container.py:91: UserWarning: Implicit dimension choice for log_softmax has been deprecated. Change the call to include dim=X as an argument.\n",
      "  input = module(input)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(tensor([ 1,  2,  3,  4,  5,  6,  7,  8,  9,  0,  1,  2,  3,  4,\n",
       "          5,  6]),\n",
       " tensor([ 1,  2,  3,  4,  5,  6,  7,  8,  9,  0,  1,  2,  3,  4,\n",
       "          5,  6]))"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "adversarial_model = deepcopy(model)\n",
    "\n",
    "residual = torch.rand(x.shape)\n",
    "residual = (residual + residual.min()) / (residual.max() * 100.)\n",
    "residual.requires_grad = True\n",
    "print(residual[0][0][0][0])\n",
    "\n",
    "pred = adversarial_model(x + residual)\n",
    "pred.argmax(1), y"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "last loss value :  3.815025806427002\n"
     ]
    }
   ],
   "source": [
    "aversarial_loss = nn.NLLLoss()\n",
    "adversarial_optimizer = optim.SGD([residual], lr=100)\n",
    "\n",
    "adv_loss_history = []\n",
    "for i in range(10000):\n",
    "    pred = adversarial_model((x + nn.Sigmoid()(residual)) / 2.)\n",
    "    adversarial_optimizer.zero_grad()\n",
    "    #adversarial loss computation\n",
    "    adv_loss  = (2 - i / 10000) * aversarial_loss(pred, torch.LongTensor([0] * 16)) #class transfert\n",
    "    adv_loss += 5 * nn.MSELoss()((x + nn.Sigmoid()(residual)) / 2., x)   #disformation should not be too big \n",
    "    adv_loss += .0001 * nn.Sigmoid()(residual).abs().sum()               #L1 regularization on the noize\n",
    "    adv_loss += .5 * nn.MSELoss()(residual, torch.zeros(residual.shape)) #L2 regularization on the noize\n",
    "    adv_loss.backward()\n",
    "    adversarial_optimizer.step()\n",
    "    adv_loss_history.append(adv_loss.item())\n",
    "    if not i % 100:\n",
    "        clear_output()\n",
    "        plt.title(\"Adversarial Loss History\")\n",
    "        plt.plot(adv_loss_history)\n",
    "        plt.grid()\n",
    "        plt.show()\n",
    "        print(\"last loss value : \", adv_loss_history[-1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/site-packages/torch/nn/modules/container.py:91: UserWarning: Implicit dimension choice for log_softmax has been deprecated. Change the call to include dim=X as an argument.\n",
      "  input = module(input)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(tensor([ 3,  6,  7,  9,  1,  1,  7,  8,  2,  6,  1,  1,  1,  1,\n",
       "          1,  9]),\n",
       " tensor([ 1,  2,  3,  4,  5,  6,  7,  8,  9,  0,  1,  2,  3,  4,\n",
       "          5,  6]))"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred = adversarial_model(x + nn.Sigmoid()(residual))\n",
    "pred.argmax(1), y"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(((x + nn.Sigmoid()(residual))[1] / 2.).detach().numpy()[0], cmap='gray')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(tensor(1.00000e-02 *\n",
       "        9.5752), tensor(0.8333))"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "nn.Sigmoid()(residual).min(), nn.Sigmoid()(residual).max()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(((x + nn.Sigmoid()(residual))[2] / 2.).detach().numpy()[0], cmap='gray')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMi4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvhp/UCwAAFCtJREFUeJzt3W2MleWZB/D/xcsAgsjLACLFRSvZBA0LZiSrNZtuum2sNgpfTE1UNjGdfijJNmniGlZc/GLIZtvKh00TqqS4VOuS1jAfdLcuMTElm8aRsAJlV1mDVgIMw/ubzgvXfpiHZtR5rv+Zc59znkPu/y8xzJxrnnPu85xzeV6u+75uc3eISH4mVD0AEamGkl8kU0p+kUwp+UUypeQXyZSSXyRTSn6RTCn5RTKl5BfJ1KRW3tiECRN80qSW3qRUrJkzSM2sadd9rRoaGsKVK1dqOjFJmWhm9wHYDGAigBfcfVN4Y5MmYd68eSk3WbfUJ0r0JG72FOkJE5r3Bo2dlytXroRxdt+j41MfE3ZeqvyfAzsv0dhSnk8nTpyo+W/rflaZ2UQA/wLg2wCWAXjEzJbVe30i0lopLymrABxy9w/dfQDArwA81JhhiUizpST/IgB/HPX7J8Vln2Nm3WbWa2a97C2kiLRO07/td/ct7t7l7l3N/OwqIuOTko1HACwe9ftXistE5BqQkvzvAFhqZreYWQeA7wLoacywRKTZ6i71ufuQma0D8B8YKfVtdfcDNRxX700mlW5SS1bR/AQ2LvZxZ2BgIIwzEydOrPu2BwcH677uWuIR9piwsbF49Likzjdh95s9n9qhg1bSGXD31wG83qCxiEgL6Rs4kUwp+UUypeQXyZSSXyRTSn6RTCn5RTLV8sX1Ue2V1X2Hh4dLYylLKGs5/uLFi3VfN4t3dHSEcebcuXOlMVaPnjFjRhifP39+GJ81a1YYT1kKzeYoDA0NhfHLly+Xxj777LOmXTfAz3s0R6FVS5H1yi+SKSW/SKaU/CKZUvKLZErJL5IpJb9Iptqqj3bK0tjUJZKsrDRt2rS6r5st2Y1KmLWYO3duaYwtXWXn/NSpU3WN6aopU6aUxqZOnRoe++mnn4bxqMQJxOedlfLYY8aOj+43kLYMu1Ht8PTKL5IpJb9IppT8IplS8otkSskvkiklv0imlPwimWqrJb2svpmy4w+ryzKTJ08ujbEW0qzWft1119V92wCwatWq0tjq1avDY+++++4wfuHChTDOavHRY7Zjx47w2O3btyfd9unTp0tjbH4Dq9Ozpcwpy7TZcuNGLfnVK79IppT8IplS8otkSskvkiklv0imlPwimVLyi2TKErfMPgzgPIBhAEPu3hX9fUdHh0etoFNaYKe27mZzCG644YbS2KVLl8Jj2RwDFn/22WfD+PLly0tj7H6z22bnla17nzlzZmmMzX/Yu3dvGO/u7g7j0WPK5lawOj2r87NafXT97PkUzSvp6+vDwMBATRMBGjHJ56/dvb8B1yMiLaS3/SKZSk1+B/BbM3vXzOL3YCLSVlLf9t/r7kfMbD6AN83sf9z97dF/UPxPoRvgWxiJSOskvfK7+5Hi3z4ArwH40goTd9/i7l3u3pWyMEdEGqvubDSz6WZ2/dWfAXwLwP5GDUxEmivlbf8CAK8VpaRJAF52939vyKhEpOnqTn53/xDAX9RxXGksZRttVjNOnQcQ9UqP5gAAwKZNm8I42wab1YznzJlTGmP3m+1HcPbs2TDO1tRH9XDWd3/FihVh/K677grjfX19pTE2P4HV+dl5Y3sxRM8ndtvRseNZ668P4SKZUvKLZErJL5IpJb9IppT8IplS8otkKmlJ73h1dHT4vHnzSuOsPBKVMVipL7UU2NVVvlp5w4YN4bFnzpwJ46xNdFTKA4CLFy+Wxth2ztOnTw/jL7zwQhhnS4LXrFlTGrv++uvDY9myW1bWuueee0pjrETJsOcLa7cePR/Z1uXR/R7Pkl698otkSskvkiklv0imlPwimVLyi2RKyS+SKSW/SKauqS26o9opO5bV0tny0mXLlpXGWM2YLdFk9ezz58+H8Y8//rg0dujQofDYbdu2hXG2RTerZx88eLA0tnnz5vBY9phGS3YB4MEHHyyNsfvN5oWwOSmXL18O49F9Y0vE2fOlVnrlF8mUkl8kU0p+kUwp+UUypeQXyZSSXyRTSn6RTLW0zm9mYV04pbcAqwmzuizbSuzYsWPjHtNVrGZ86tSpML5z584wvmPHjtIYa1HN5j9EvQIAXuc/cOBAaYydF7ZVNauHR/eN1cpZr4GUVu9A3I6dPWaNyiG98otkSskvkiklv0imlPwimVLyi2RKyS+SKSW/SKZond/MtgL4DoA+d7+juGwOgFcBLAFwGMDD7n6aXdfw8HC4Np3V4qP65uzZs8NjWd2V1Yx7enpKY2yOAeu7v2vXrjDOxn7LLbeUxljN+KOPPgrjqfsh3H777aUx1iuA1dLZYx71YFi5cmV4LOvvwHo0sHp7f39/aYz1AhgcHKz7dker5ZX/FwDu+8JlTwHY5e5LAewqfheRawhNfnd/G8AXp6A9BOBqK5RtAFY3eFwi0mT1fuZf4O5Hi5+PAVjQoPGISIskz+13dzez0g8aZtYNoBvgn41FpHXqzcbjZrYQAIp/SzspuvsWd+9y9y72BY6ItE69yd8DYG3x81oA8bIzEWk7NPnN7BUA/wXgz83sEzN7AsAmAN80sw8A/E3xu4hcQ+hnfnd/pCT0jfHe2MSJEzFjxozSOKvzR/FofXQt13327NkwHs0D2L9/f3js/Pnzw/jMmTPD+JkzZ+qOszkCbD0/W/d+8uTJMP7oo4+Wxth+B0NDQ2GczWGI5k+wnv+sTwGbB8COjx4X9t1YNMdgPB+t9Q2cSKaU/CKZUvKLZErJL5IpJb9IppT8Iplqaetudw/LM6x9dlTGYEsZp06dGsY7OzvDeISVCVn5hZW82BLP6L6zchlbmprSRhoAVq1aVRpj542VKXt7e8P47t27S2PTpk0Lj2XY2Nh5i0reKS3sVeoTEUrJL5IpJb9IppT8IplS8otkSskvkiklv0imWr5Fd1RXZrX406fLu4OzuitrMR21FAfiLZtZrZwtyWX3m9Vuo3hUTwb4kt2oxTQAPP/882E8Ou9sG2w2tg0bNtR9fGpLOVaLZ8+3aB5AyuM9HnrlF8mUkl8kU0p+kUwp+UUypeQXyZSSXyRTSn6RTLW0zg/E9VHWfjuqp7Nj2Zr4RYsWhfGoVp/aNnzWrFlhnK09j+KsV8CxY8fC+Nq1a8P4kiVLwnjUIpv1Anj//ffD+Lx588J4NPfj0qVL4bHsMWN1fDaPIBobu+1GbXunV36RTCn5RTKl5BfJlJJfJFNKfpFMKflFMqXkF8kUrfOb2VYA3wHQ5+53FJdtBPA9ACeKP1vv7q+z63L3sI98yjpm1vN/9uzZYZzVTm+88ca6b3vv3r1hnPUiWL58eRg/fvx4XTEAeOyxx5LiKX0Q2Hl58sknwzirtads6Z5Spwd4j4bBwcHSGMuDaH5Eo/v2/wLAfWNc/lN3X1H8RxNfRNoLTX53fxvAqRaMRURaKOUz/zoze8/MtppZ/J5aRNpOvcn/MwBfBbACwFEAPy77QzPrNrNeM+tln5NEpHXqSn53P+7uw+5+BcDPAZTuxujuW9y9y927GrUgQUTS1ZWNZrZw1K9rAOxvzHBEpFVqKfW9AuDrADrN7BMA/wjg62a2AoADOAzg+00co4g0AU1+d39kjItfrPcGo7f+bI11VNdlfdTZPvXTp08P41EP+IsXL4bHLl68OIx3dnaG8blz54bx6LytWbMmPPaBBx4I46yOz+rZ0XnbuXNneOzSpUvDOHu+RPs8sOcLm7vB5hikxNl6/ui5zO7XaPoQLpIpJb9IppT8IplS8otkSskvkiklv0imbDylgVQdHR0etVtmyxGjEggrt7HyCdvKOhrbbbfdFh67cOHCMM5aWEftrwHg8ccfL43deeed4bFMtPQU4CWxZ555pjS2Z8+e8Fi29Tk7b5H58+eHcVYaZs/VqMwI8JbqkagNfX9/PwYGBmpa16tXfpFMKflFMqXkF8mUkl8kU0p+kUwp+UUypeQXyVRb1flTxsJaMbOlqayuGy3BZMtab7311jDO6vhPP/10GF+5cmVpjNXho9baQHoL66hl+u7du8NjWTxl2Sw7dt++fWGctR1ncxQGBgZKY+y5GM29OHnyJAYHB1XnF5FySn6RTCn5RTKl5BfJlJJfJFNKfpFMKflFMtXSOv/kyZN9zpw5pXE2lqh2yrbgZtj666i195kzZ8JjWU35ueeeC+OsH8DNN99c921H9WaA9zlg7bPPnTtX921PmTIljLPnS9Q2nJ2Xs2fPhvFXX301jO/YsSOMR/Mn2P2K5lb09fVpPb+IxJT8IplS8otkSskvkiklv0imlPwimVLyi2SKbtFtZosBvARgAQAHsMXdN5vZHACvAlgC4DCAh909LJabWVirZ2vPo3XM/f394bGsTzvrvR/VXlm9et26dWE8qtMDwLRp08J4ythYX/6oTg/wenkKthcDmwcQPZ/Y3AyGrbln/SWix4ztR8D2DKhVLa/8QwB+5O7LAPwlgB+Y2TIATwHY5e5LAewqfheRawRNfnc/6u57ip/PAzgIYBGAhwBsK/5sG4DVzRqkiDTeuD7zm9kSACsB/B7AAnc/WoSOYeRjgYhcI2pOfjObAeDXAH7o7p/7IOgjH2DG/BBjZt1m1mtmvazfm4i0Tk3Jb2aTMZL4v3T33xQXHzezhUV8IYAxu1C6+xZ373L3LtYMUkRah2ajjXy1+CKAg+7+k1GhHgBri5/XAtjZ+OGJSLPUUqf5GoDHAOwzs6v9itcD2ATg38zsCQAfAXi4lhuMShys7BRta8xKWmyZJFvCGX1kYUtyb7rppjDOWn/PnTs3jEeloWg7Z4BvFc3Kaaws9cYbb5TGWHtrNjZWTovKt1G7c4C3JGe3zZ6PUYmU3Xaj3kHT5Hf33wEoKyx+oyGjEJGW04dwkUwp+UUypeQXyZSSXyRTSn6RTCn5RTLV8tbds2bNKo2z+ma0RJPVPtlW1GzpajS2t956KzyWLU1lS3bZ2KOaMtua/OWXXw7jrI6/ffv2MB7NYWBtwaPW2wCfu3HhwoXSGDsvbI7B8PBwGGfnrVmzXU+cOKHW3SISU/KLZErJL5IpJb9IppT8IplS8otkSskvkqnm9V0uEdU3WRvoKM7qpmwOAVtTH7Vq7unpCY9dv359GGdj27hxYxjfunVraYy1eU5tvc3mIETnja15Z3MvWK+BlP4P7LyxXgTsvEZzVlhb8EbNzdErv0imlPwimVLyi2RKyS+SKSW/SKaU/CKZUvKLZKql6/k7Ojq8s7OzfDCktpqy3RebB5ByHti4U7dcvnTpUhiPatbstlm9ms1/YOvaWc26WccC8fOF1eFTny/NfD5F+vr6tJ5fRGJKfpFMKflFMqXkF8mUkl8kU0p+kUwp+UUyRRdzm9liAC8BWADAAWxx981mthHA9wCcKP50vbu/XsP1lcZYbTSqvabWXVk8Wn/NasZsfgKr47O6b8o5ZXV6tg99Sv/56JzWImUf+5Q5I+y6a7n+lMesUWrp5DAE4EfuvsfMrgfwrpm9WcR+6u7/3LzhiUiz0OR396MAjhY/nzezgwAWNXtgItJc43rPZmZLAKwE8PvionVm9p6ZbTWz2SXHdJtZr5n1pr7VEpHGqTn5zWwGgF8D+KG7nwPwMwBfBbACI+8MfjzWce6+xd273L2rWfuTicj41ZSNZjYZI4n/S3f/DQC4+3F3H3b3KwB+DmBV84YpIo1Gk99GvpZ8EcBBd//JqMsXjvqzNQD2N354ItIstXzb/zUAjwHYZ2Z7i8vWA3jEzFZgpPx3GMD3UwfDSlpRCSRlGSTAl75G31ewNtCsnMZKXs1uvx1hY0/5HoddNyt5pTzmqR9Bmzm2Vqnl2/7fARjrntCavoi0L30DJ5IpJb9IppT8IplS8otkSskvkiklv0imWr5Fd7ti9eqUOQapS35T2kg3sx06u+1UKfM+gLQl4Kl1+tSxN+vY0fTKL5IpJb9IppT8IplS8otkSskvkiklv0imlPwimWrpFt1mdgLAR6Mu6gTQ37IBjE+7jq1dxwVobPVq5Nj+zN3n1fKHLU3+L924Wa+7d1U2gEC7jq1dxwVobPWqamx62y+SKSW/SKaqTv4tFd9+pF3H1q7jAjS2elUytko/84tIdap+5ReRilSS/GZ2n5n9r5kdMrOnqhhDGTM7bGb7zGyvmfVWPJatZtZnZvtHXTbHzN40sw+Kf8fcJq2isW00syPFudtrZvdXNLbFZvaWmf3BzA6Y2d8Vl1d67oJxVXLeWv6238wmAngfwDcBfALgHQCPuPsfWjqQEmZ2GECXu1deEzazvwJwAcBL7n5Hcdk/ATjl7puK/3HOdve/b5OxbQRwoeqdm4sNZRaO3lkawGoAf4sKz10wrodRwXmr4pV/FYBD7v6huw8A+BWAhyoYR9tz97cBnPrCxQ8B2Fb8vA0jT56WKxlbW3D3o+6+p/j5PICrO0tXeu6CcVWiiuRfBOCPo37/BO215bcD+K2ZvWtm3VUPZgwLim3TAeAYgAVVDmYMdOfmVvrCztJtc+7q2fG60fSF35fd6+53Avg2gB8Ub2/bko98Zmunck1NOze3yhg7S/9Jleeu3h2vG62K5D8CYPGo379SXNYW3P1I8W8fgNfQfrsPH7+6SWrxb1/F4/mTdtq5eaydpdEG566ddryuIvnfAbDUzG4xsw4A3wXQU8E4vsTMphdfxMDMpgP4Ftpv9+EeAGuLn9cC2FnhWD6nXXZuLttZGhWfu7bb8drdW/4fgPsx8o3//wH4hyrGUDKuWwH8d/HfgarHBuAVjLwNHMTIdyNPAJgLYBeADwD8J4A5bTS2fwWwD8B7GEm0hRWN7V6MvKV/D8De4r/7qz53wbgqOW+a4SeSKX3hJ5IpJb9IppT8IplS8otkSskvkiklv0imlPwimVLyi2Tq/wHYk8MelTs3pAAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMi4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvhp/UCwAAFGxJREFUeJzt3X2M1eWVB/DvAWYEZsDhZUBiWUBEDNFAzYCaJcrq2lhTov3HYOKGJg3TxKLb2MQlGl8S/jHrYqNk0zhdsbhhbTdpDcb4AoubuJUVBbQoZQGXUIXMKyBvgsPL2T/mRzPi/M65c58793fZ8/0kZGbumd+9z/zuPdyX85znEVUFEcUzrOgBEFExmPxEQTH5iYJi8hMFxeQnCorJTxQUk58oKCY/UVBMfqKgRlTzxoYNG6YjRlT1Jks2lDMdRWTIrpuov7Nnz+L8+fMlPeCSMlFE7gTwHIDhAP5FVZ82b2zECDQ3N6fcZK7U5D1z5kzZxw4bZr+AGj58eNLx58+fH/SYLkj9j8cbm3feU+4X79iUv827T1KdO3fOjFt/W8rf1d3dXfLvlv2yX0SGA/hnAN8HMAfAfSIyp9zrI6LqSnnPvwDAZ6q6T1V7AfwGwN2VGRYRDbWU5L8SwBf9fj6QXfYNItIqIltFZGvKy1ciqqwh/7RfVdtUtUVVW7z3j0RUPSnZeBDA1H4/fye7jIguASnJ/yGAWSIyQ0TqASwB8FplhkVEQ63sUp+qnhWR5QDeRl+pb42q7qzYyAYptXTjzT+wPq84e/aseaxX9vHG7pV+rHhqmdGLnzp1qux4ann2sssuM+PW2FPPi3efeNdvlZartbpWUp1fVd8A8EaFxkJEVcRP4IiCYvITBcXkJwqKyU8UFJOfKCgmP1FQtdlcX4bUvgGvzm/Vbb1at1fn9+YJ1NXVmXFr7N558eJePdurtTc0NOTGUlt2vTbsMWPGmHHLV199ZcZHjhyZdLylWnV+PvMTBcXkJwqKyU8UFJOfKCgmP1FQTH6ioGqq1Nfb22vGrZKYVxZKadkF7HLb2LFjzWNPnz6dFPdKgU1NTbmxr7/+2jzWK1Na1w0A48aNM+PW2Nvb281jvbbakydPmvGOjo7cmHd/e4+XUaNGmXFPfX19bsxrB65UKZDP/ERBMfmJgmLyEwXF5CcKislPFBSTnygoJj9RUDVV5/faQxsbG3NjXq3ca//02matenXK/ATArynPnz/fjF977bW5sWXLlpnHejXjl156yYxv27bNjB85ciQ35rXcdnV1mfGjR4+aca+V2uLNG/HmT4wePbrs62dLLxENKSY/UVBMfqKgmPxEQTH5iYJi8hMFxeQnCkpSaooish/AcQDnAJxV1Rbr9+vq6tTq//ZqqxarPxrwe8O9eQLWUsxenX769Olm3Kvj33LLLWb8hhtuyI15/fheP/+hQ4fMuNdTv3LlytzY7t27zWO95a+9+9SaN5LyWAP8WrwXtx4z3rwQ6+/u7u5Gb29vSX9cJSb5/I2q9lTgeoioiviynyio1ORXABtEZJuItFZiQERUHakv+xeq6kERmQRgo4j8j6q+2/8Xsv8UWgH/PRoRVU9SNqrqwexrF4BXASwY4HfaVLVFVVuY/ES1o+xsFJEGERlz4XsA3wPwaaUGRkRDK+Vl/2QAr2YlkxEA/k1V36rIqIhoyJWd/Kq6D8DcwRwjImbt1VtL3aqder3bQ7kV9erVq81jp02bZsYnTpxoxr2+d2stAq9m7K2h0NzcbMa9vvWFCxfmxj744IOk2/bOm/U288SJE+axKfM+AH8Lb+9+qQa+CScKislPFBSTnygoJj9RUEx+oqCY/ERBJbX0DlZ9fb1OmjQpN+7NALTiXmtqynUDwNy5+VXNdevWmcd65TCPt+y4Vcb0zktDQ4MZP3bsmBn3SoVWq/WiRYvMY7u7u824d59Z5V+vDdvjLd3tLQXvlRIt1hbePT09Jbf08pmfKCgmP1FQTH6ioJj8REEx+YmCYvITBcXkJwqqprbo9li1Va+FcuzYsWbca7GcPXt2bsybK/H555+b8SlTppjxnh57ceQNGzbkxtrb281jFy9ebMateRmlXL+1bPmKFSvMY1etWpV029YW3l4LuFent2rtgN/ya81/8B7LKVuP98dnfqKgmPxEQTH5iYJi8hMFxeQnCorJTxQUk58oqKrX+a0ebG95basv3uvtHj9+fNnjAoCNGzfmxm6//Xbz2FmzZpnxhx9+2Izv2LHDjO/bty835vWd79y504w//vjjZtyzd+/e3FhbW5t5rPV3Af7fZs2/8O7vlMci4K8XYF2/N4fAWmZ+MFuP85mfKCgmP1FQTH6ioJj8REEx+YmCYvITBcXkJwrKrfOLyBoAPwDQparXZZeNB/BbANMB7Adwr6oe8a5LVc016L0apdXH7G257K0/7/VQW/3dy5cvL/tYwK8Je2sNWGvvezXjhx56yIx79W7v+mfMmJEbmzp1qnnsF198YcY91nnz+u29eQCNjY1m/OTJk2bc2u8gJQ8Gsw9HKc/8vwZw50WXrQCwSVVnAdiU/UxElxA3+VX1XQCHL7r4bgBrs+/XArinwuMioiFW7nv+yap6YQ2lDgCTKzQeIqqS5Ln9qqoikvtGQ0RaAbQC/vsoIqqecrOxU0SmAED2tSvvF1W1TVVbVLWFyU9UO8rNxtcALM2+XwpgfWWGQ0TV4ia/iLwC4L8BzBaRAyLyYwBPA7hDRPYC+NvsZyK6hMhg6oKp6urqdOLEiblxbywpe6qn9EgDQHNzc26st7fXPDa1Juz1rVv17IULF5rHPvHEE2bcOy/efgjW+vTe3AxvnYPNmzebccupU6fKPhbwH4ve3A7rvHrzOqy5F93d3ejt7S2pqZ9vwomCYvITBcXkJwqKyU8UFJOfKCgmP1FQVS/1NTU15ca9sVjlEa/04pXbvNueMGFCbix15uLx48fN+OHDF/dVfZP1t3stuc8//7wZv+aaa8y4t130mDFjzLjFav8GgMcee8yMv/POO7kxq6W2lLhXAvVKyynbbFvHdnR0sNRHRDYmP1FQTH6ioJj8REEx+YmCYvITBcXkJwqqqnX++vp6nTRpUm48pc7v1V2t1lLAb6O0WkAvv/zypNv2auFHjtiroh86dCg35rUb33rrrWb8/vvvN+NWizZgt/x6ba9erd27z1pbW3Nje/bsMY/17jNvqXePNXYvD6x5JazzE5GLyU8UFJOfKCgmP1FQTH6ioJj8REEx+YmCSt6ua7CserzXFz9q1KjcmNe3bvXjA36/v1VL95aB9pbe9raLHj9+vBm3+vm9Ov+uXbvM+MqVK834zJkzzfiDDz6YG7v66qvNY0ePHm3GvfkP48aNy41595m3xoL3WPX6+a15At4cg0rNzeEzP1FQTH6ioJj8REEx+YmCYvITBcXkJwqKyU8UlNvPLyJrAPwAQJeqXpdd9hSAZQC6s197VFXf8G6srq5OU9a/b2hoyI15/dVe3dXrLbfmEXhzBLw6/9GjR824t16Adby3DbY3du8+8eYRWPf3m2++aR7r1fG9+3z79u25MW+dAm9dfW+fiJR1/b11DKyx9fT04MyZMxXr5/81gDsHuPwXqjov++cmPhHVFjf5VfVdAPaWMUR0yUl5z79cRHaIyBoRyZ9HSUQ1qdzk/yWAmQDmAWgHsCrvF0WkVUS2ishWb/49EVVPWcmvqp2qek5VzwP4FYAFxu+2qWqLqrakbmhJRJVTVjaKyJR+P/4QwKeVGQ4RVYvb0isirwBYBGCiiBwA8CSARSIyD4AC2A/gJ0M4RiIaAm7yq+p9A1z8Yjk3NmzYMLM269WMLV4926uVe29Jenp6cmMnT540j/X69b34sWPHzHhzc3NuzKvjp/L2HLBq0t591tTUZMa9+RNz5szJjd1xxx3msVu2bDHj3mPVezxZPfunT582j/XmEJSKb8KJgmLyEwXF5CcKislPFBSTnygoJj9RUFVdultVza2JvbZaq7zitVh65TgvbpWVvJKVt421d7y1zTVgj807p94S1dZy6YC/TfaXX35Z9rFey643duu8e6Vfj9VeDvgtwdZj2ZsGbz3WB1MG5DM/UVBMfqKgmPxEQTH5iYJi8hMFxeQnCorJTxRU1bfotlodvTZIq3bqtb2eOXOm7HEBdtvtbbfdZh5rtZYCwPr16834pEmTzLg1dm8ZaK819eDBg2bc207a2ibbm4Nw+LC9bqy3HLvVhm3NPwD8bbC9eSHe32bV4705K5XCZ36ioJj8REEx+YmCYvITBcXkJwqKyU8UFJOfKKiq1/ktXi3e4i137NWzvb711atX58YWL15sHuvVfOfOnWvGX3jhBTNu1bu9nveOjg4z7i3NPX/+fDP+yCOP5MZOnTplHuvNIfDq/G+99VZu7KOPPjKP9frxU+akAHbPfur236XiMz9RUEx+oqCY/ERBMfmJgmLyEwXF5CcKislPFJR4fcsiMhXAywAmA1AAbar6nIiMB/BbANMB7Adwr6oesa6rvr5ere2kvbFY8wC8tc692uiECRPM+Pvvv58bs3rWAWDPnj1m3JujsHnzZjP++uuv58a8OQbTpk0z4w888IAZHz16tBm35k946/Z7W3AfOWI+3Myx79u3zzzWq7V7dX7v8WgpISdzY52dnejt7S1pIkApz/xnAfxcVecAuAnAT0VkDoAVADap6iwAm7KfiegS4Sa/qrar6vbs++MAdgG4EsDdANZmv7YWwD1DNUgiqrxBvecXkekAvgtgC4DJqtqehTrQ97aAiC4RJSe/iDQC+B2An6nqNxbM0743KQO+URGRVhHZKiJbU94HEVFllZT8IlKHvsRfp6q/zy7uFJEpWXwKgK6BjlXVNlVtUdUW70MSIqoeNxul76PFFwHsUtVn+4VeA7A0+34pAHsJWiKqKaWU+hYC+C8AnwC48Lr9UfS97/93AH8F4M/oK/WZay3X19ertW1yytbE3rFe2cg7DzfddFNu7MknnzSPveKKK8z40aNHzbjXVmstIz19+nTzWG9pb29JdK9cl/JW79ChQ2b82WefNeObNm3KjXntwl57ufd3V6rtdiDWOe3u7i651Of286vqHwDkXdntpdwIEdUevgknCorJTxQUk58oKCY/UVBMfqKgmPxEQbl1/kry6vwpUlt6veOt+M0332weu2TJEjN+/fXXm3Gvbfaqq67KjXntxN7978UbGxvNuHXevGXFn3nmGTO+e/duM24tS+61OntLvXuPF++8pcx2tW67q6uroi29RPT/EJOfKCgmP1FQTH6ioJj8REEx+YmCYvITBVX1Or+1dPdQLvOVWne1tlz25hCMHDnSjHtbfM+ePduM33jjjbkxb+vxlG3RAeDtt98249by2u+99555rLcFd2dnpxm3tgD3+vG9207Nm5R+f9b5iSgJk58oKCY/UVBMfqKgmPxEQTH5iYJi8hMFdUnV+a1avHdsat3WqvOnjBvwe8u9608Zm8c7b14/f1NTU27M6+c/ceKEGffmKFi1dO8+8bboLrKf37pu1vmJyMXkJwqKyU8UFJOfKCgmP1FQTH6ioJj8REG5W3SLyFQALwOYDEABtKnqcyLyFIBlALqzX31UVd/wrs+qUaasre8dm7o+fUpd1qsZp861sOYJeOP21hrwjm9oaDDjVj+/d59Z8xcA4PTp02bcWssgdZ8H7z5L6ddP3WOiVG7yAzgL4Oequl1ExgDYJiIbs9gvVPWfKjISIqoqN/lVtR1Ae/b9cRHZBeDKoR4YEQ2tQb2WFZHpAL4LYEt20XIR2SEia0RkXM4xrSKyVUS2DuUyXUQ0OCUnv4g0AvgdgJ+p6jEAvwQwE8A89L0yWDXQcarapqotqtqS8r6ZiCqrpGwUkTr0Jf46Vf09AKhqp6qeU9XzAH4FYMHQDZOIKs1Nfun76PFFALtU9dl+l0/p92s/BPBp5YdHREOllE/7/xrA3wH4REQ+zi57FMB9IjIPfeW//QB+MiQj7McqgaSUVoC0Up/X9urxPgvxWn6tUqK3RLW1vHUpvHKbNbbUbbC9rctTSsNemTFVytgqpZRP+/8AYKDRuDV9Iqpd/ASOKCgmP1FQTH6ioJj8REEx+YmCYvITBVVKnb+iqlXDvFhqy25KK7IntX3Uq7Wn8OrdXruyV8tPue7UsVlSW35Tjk+97VLxmZ8oKCY/UVBMfqKgmPxEQTH5iYJi8hMFxeQnCqqqW3SLSDeAP/e7aCKAnqoNYHBqdWy1Oi6AYytXJcc2TVWbS/nFqib/t25cZKuqthQ2AEOtjq1WxwVwbOUqamx82U8UFJOfKKiik7+t4Nu31OrYanVcAMdWrkLGVuh7fiIqTtHP/ERUkEKSX0TuFJHdIvKZiKwoYgx5RGS/iHwiIh+LyNaCx7JGRLpE5NN+l40XkY0isjf7OuA2aQWN7SkROZidu49F5K6CxjZVRP5TRP4kIjtF5O+zyws9d8a4CjlvVX/ZLyLDAewBcAeAAwA+BHCfqv6pqgPJISL7AbSoauE1YRG5BcAJAC+r6nXZZf8I4LCqPp39xzlOVf+hRsb2FIATRe/cnG0oM6X/ztIA7gHwIxR47oxx3YsCzlsRz/wLAHymqvtUtRfAbwDcXcA4ap6qvgvg8EUX3w1gbfb9WvQ9eKouZ2w1QVXbVXV79v1xABd2li703BnjKkQRyX8lgC/6/XwAtbXltwLYICLbRKS16MEMYHK2bToAdACYXORgBuDu3FxNF+0sXTPnrpwdryuNH/h920JVvQHA9wH8NHt5W5O07z1bLZVrStq5uVoG2Fn6L4o8d+XueF1pRST/QQBT+/38neyymqCqB7OvXQBeRe3tPtx5YZPU7GtXweP5i1rauXmgnaVRA+eulna8LiL5PwQwS0RmiEg9gCUAXitgHN8iIg3ZBzEQkQYA30Pt7T78GoCl2fdLAawvcCzfUCs7N+ftLI2Cz13N7XitqlX/B+Au9H3i/78AHitiDDnjugrAH7N/O4seG4BX0Pcy8Az6Phv5MYAJADYB2AvgPwCMr6Gx/SuATwDsQF+iTSlobAvR95J+B4CPs393FX3ujHEVct44w48oKH7gRxQUk58oKCY/UVBMfqKgmPxEQTH5iYJi8hMFxeQnCur/ALTa1vKKPCu/AAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMi4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvhp/UCwAAE3hJREFUeJzt3V2MlGWWB/D/oWmQ70ZB0jLsOhCyiRGXWVuC+JExs4NCiMiFMsZMmDiZnouB7CRzsUQvFmM0ZuOAXmwm6XHI4IZlxjhjxMTMDks2YUdXtCF+4u6CpMnQaWm0raYbFLqbsxf1Mimx65zqeqrqLTz/X0K6u049VU+9bx3q4zwfoqogongm5d0BIsoHk58oKCY/UVBMfqKgmPxEQTH5iYJi8hMFxeQnCorJTxTU5Ebe2aRJk7SlpaWRd0lfY97oVBH5Wt63ZWxsDBcvXqzozpOSX0TuAfAsgBYAz6nqU9b1W1pa0NbWlnKXVl/MeOrJyutkAsDFixerbpv6n6133FLjKZj8X1UoFCq+btVv+0WkBcC/AFgD4AYAD4rIDdXeHhE1Vspn/hUAjqnqcVW9AOA3ANbXpltEVG8pyb8QwJ9L/j6ZXfYlItIpIt0i0p3y9pWIaqvu3/arapeqdqhqx6RJLC4QNYuUbOwFsKjk729klxHRFSAl+d8CsFREvikiUwB8D8De2nSLiOqt6lKfqo6KyGYA/45iqW+nqn5Qs55NvD9J7UdGRsy4Vbrx7ru1tbXq266E1b7e37OklFhTy6sp59xrW++PqCnnpVZlxKQ6v6q+CuDVmvSEiBqK38ARBcXkJwqKyU8UFJOfKCgmP1FQTH6ioBo6nz+VVZtNrfN7ddfR0dGysXPnziXd9+TJ9mmYPXt21fHz589X1adLvOOSUov3aunefXvn3JrOnHrbnpTbr+f4hVJ85ScKislPFBSTnygoJj9RUEx+oqCY/ERBXVGlPqus5JWcxsbGzLi3yq1VuvFKcV7fLly4YMbPnj1rxq3pyF4Z0YunTgm2ynneNOrUMqMV954PqdOJvb5bce+5WKvpxnzlJwqKyU8UFJOfKCgmP1FQTH6ioJj8REEx+YmCaqo6f8qOr17d1rvtKVOmmPGrrrqqbGzq1KlmW88XX3xhxr3HZk0p9tp69Whv2fGU8RPeMa/nDsHeOUsZ9wHYU8CBtGm7KeNdSvGVnygoJj9RUEx+oqCY/ERBMfmJgmLyEwXF5CcKKqnOLyI9AIYAjAEYVdWOWnTKuL+qYkD6HGir7uvNifdYYwiA9GWkLV49enh42Ix79W6rnu7V0mfNmlX1bQP2+Adv/II39sJrnzI+whs7YS3HPpHnSi0G+dylqp/U4HaIqIH4tp8oqNTkVwB/FJFDItJZiw4RUWOkvu2/XVV7ReRaAPtE5H9U9UDpFbL/FDqB2q09RkTpkrJRVXuzn/0AXgKwYpzrdKlqh6p2MPmJmkfV2SgiM0Rk1qXfAawG8H6tOkZE9ZXytn8BgJeyEttkAP+mqn+oSa+IqO6qTn5VPQ7gb2vYl0rus2zMqxl7c8c91hrz3n17tXRvnIBXM547d27ZmFeH9+rZqXsOWMfNq2d78cHBQTNu9d3ba8E7p9OnTzfj3nG1avXex2MrDzifn4hcTH6ioJj8REEx+YmCYvITBcXkJwqqoUt3q2rS9FSrJOaVR7yyUYqhoSEznrrMs9feemzz5s0z27a1tZlxr/0nn9gTOq2+eY/LK7fNnz/fjFtlL6/0W88lywF7GrdXwrTKjGfOnDHbluIrP1FQTH6ioJj8REEx+YmCYvITBcXkJwqKyU8UVEPr/CIyoSmHl7Pqwt74gZRpkoBd1/3888/Nth6vpmxNiwXsmrFXb7amlgLAzTffbMYPHz5sxq0xDN60V2/arNd3awyDd0ytZb8BYNWqVWbcY53z48ePm22tY+6NnSjFV36ioJj8REEx+YmCYvITBcXkJwqKyU8UFJOfKKiG1vk9Xq3dqmF69c2UOn7qfXtbcFtLbwPAwMCAGd+yZUvZ2Jo1a8y2M2fONOPeuIz9+/eb8ddee61szNuC21srYPHixWb8pptuKhvz6vjLli0z49459R6b1d57Pt16661mvFJ85ScKislPFBSTnygoJj9RUEx+oqCY/ERBMfmJgnLr/CKyE8A6AP2qemN22dUAfgvgegA9AB5Q1c/q18103jbYXm3Vau+NEUjd5nrt2rVm/OGHHy4b89YamMj87/HcdtttZryjo6NszDsnnpQ9IDzTpk0z49459dZoOHv2bNnYjh07zLYp61qUquSV/9cA7rnssq0A9qvqUgD7s7+J6AriJr+qHgBw+RCz9QB2Zb/vAnBfjftFRHVW7Wf+Baral/3+MYAFNeoPETVI8th+VVURKftBQ0Q6AXQC/jp6RNQ41WbjKRFpB4DsZ3+5K6pql6p2qGoHk5+oeVSbjXsBbMp+3wTg5dp0h4gaxU1+EdkD4L8B/I2InBSRHwJ4CsB3ReQogL/P/iaiK4j7mV9VHywT+k6N++Ky9lT36tleLd2rOVv37e317q0v7/Xda2+tze+tFdDb22vGvXr1NddcY8ZPnz5txi3e4z527JgZX7JkSdnYkSNHzLbDw8Nm3FtrwFrHALDr/CdOnDDbWud7Ivti8EM4UVBMfqKgmPxEQTH5iYJi8hMFxeQnCqqplu62tnMG7HKdt+Wyt1W1VyKx4nPmzDHbDg0NmfHrrrvOjD/55JNm3Nrq2tsGu7+/7OBMAH657eDBg2a8u7u7bKyvr69sDPCXx7bKZYDdd+/5kjIlF/Cfy5ZGjYTlKz9RUEx+oqCY/ERBMfmJgmLyEwXF5CcKislPFFRT1fm92qpVO/Xq9Km1U2tJZG8LbW957IULF5pxb8qwtd20N+V25cqVZvyxxx4z4970U6ue7i2P7Y2PSKnze+fks8/SVqL3zpllItNyU/CVnygoJj9RUEx+oqCY/ERBMfmJgmLyEwXF5CcKSuq5zfHlWltbta2trer2KVs6e4/TmzturSXgjU8oFApm3HtcL7zwghlfvnx52Zi3xLRX7/a2H7/22mvN+ObNm8vGDhw4YLb1lh33xgFY4x883joIKes/APY5nzp1qtnWMjAwgJGRkYoGCvCVnygoJj9RUEx+oqCY/ERBMfmJgmLyEwXF5CcKyi2ci8hOAOsA9Kvqjdll2wD8CMCl/ZcfUdVXvdtSVbeuXC2vrurV4r1tsq1+e/PSp0+fbsa97cO3bt1qxu+///6ysaVLl5ptV61aZca9Or+3loG1VoH3uNvb2824N2bEGuMwY8YMs623rr/3uM+cOWPGreOaup18pSp55f81gHvGuXyHqi7P/rmJT0TNxU1+VT0AwP5vjoiuOCmf+TeLyLsislNE7HGYRNR0qk3+XwBYAmA5gD4APy93RRHpFJFuEelu5DwCIrJVlfyqekpVx1T1IoBfAlhhXLdLVTtUtaNRCxMSka+q5BeR0q9hNwB4vzbdIaJGqaTUtwfAtwHME5GTAP4JwLdFZDkABdAD4Md17CMR1UHD5/Nb68h7HwusOdZe7dOLe3V+qy47b948s+2+ffvM+Pbt2834nj17zLh1Dr1xFd66/k888YQZv+uuu8y4NQZi48aNZluv1u6t2//RRx+ZcUvq82lwcNCMW8+nlpYWs601379QKGB0dJTz+YmoPCY/UVBMfqKgmPxEQTH5iYJi8hMF1fAtulOm9FrbbFvbdwPA8PBw1fcL2NNy7733XrOtV7LasGGDGX/xxRfNuFUC9Zag9kqcn376qRn3ls+2puWuXr3abLtt2zYznjJi1Juq7E0B9+7b26LbKud5fbPaTuSY8JWfKCgmP1FQTH6ioJj8REEx+YmCYvITBcXkJwqqoVN6J0+enLRFtzVGwKvzp9ROAXva7Ouvv262nTNnjhl/6KGHzPg777xjxq3ppd4YA2vZbwB45plnzPiJEyfMuHVeVq5cabb1au3WuA8AmD17dtmYtyX7+fPnzXitls8ejzc2w1rae3BwkFN6icjG5CcKislPFBSTnygoJj9RUEx+oqCY/ERBNXQ+v4gkzclPWQvAq8t6NWNrnvTRo0fNtosXLzbjt9xyixn3avXW/VvLPAP+fP3+/n4z7s0ft9ZB8I6Lt821d86sx+7V8b1j7vHaW891b+xNrXa+4is/UVBMfqKgmPxEQTH5iYJi8hMFxeQnCorJTxSUO59fRBYBeB7AAgAKoEtVnxWRqwH8FsD1AHoAPKCqn1m35c3n9+r4Vn3Tm/udWju11qe/4447zLaPPvqoGffWeB8YGDDjr7zyStnY/PnzzbbengHeOAFvbMabb75ZNvb000+bbT3engHnzp0rG/Oea17cWx/Ce75Z59wbk2I9VwuFAkZGRmo2n38UwM9U9QYAKwH8RERuALAVwH5VXQpgf/Y3EV0h3ORX1T5VPZz9PgTgQwALAawHsCu72i4A99Wrk0RUexP6zC8i1wP4FoCDABaoal8W+hjFjwVEdIWoOPlFZCaA3wH4qap+adC1Fj/gjPshR0Q6RaRbRLobuV4gEdkqSn4RaUUx8Xer6u+zi0+JSHsWbwcw7gwQVe1S1Q5V7ajVhAQiSucmvxQz9lcAPlTV7SWhvQA2Zb9vAvBy7btHRPVSSanvdgD/BeA9AJfqH4+g+Ln/BQB/BeAEiqU+sybV2tqqc+fOrbqzVl+9pbe9UqBXsioUCmbcsmXLFjO+ceNGM+5to209Nq9s5J1/b9rsG2+8YcYff/zxsjFv+Wwv7rGWwPbKq9by2ED6lu/WlF+vvGqVISdS6nPn86vqnwCUu7HvVHInRNR8OMKPKCgmP1FQTH6ioJj8REEx+YmCYvITBdXQLbq9On/KtFtv9KC37bE3hdOqOXs1Y28Mwt13323G77zzTjO+bNmysjFrm2rAP+bPPfecGd+9e7cZt7Yn92rlXq3dWx7bmnbrjSHwxjd4S397z0dvSnC1CoUCt+gmIhuTnygoJj9RUEx+oqCY/ERBMfmJgmLyEwXV8Dq/tXS3Vxu1avFeW69m7NV9rfn+Xh1/1qxZZjy1b+vWrSsb887voUOHzHhPT48Z98ZHWMfGa+tt0e3V2lPWf/Dm1HvjAFK2fE8Z71LrpbuJ6GuIyU8UFJOfKCgmP1FQTH6ioJj8REEx+YmCaqo6f0pfUudPp+wm5NV0Palr53t7ElhS5sQDfq3e2jcgdS8Fr+/WcU3d0t0bJ5Ai5bnKOj8RuZj8REEx+YmCYvITBcXkJwqKyU8UFJOfKCh3i24RWQTgeQALACiALlV9VkS2AfgRgNPZVR9R1VdTOpNSa/d4dVmvlm7VnFPr0R6rVg7YdWGvXu3V6b09CbxzZt1/6n1PmzbNjFvnxVtDIeVxVRK3bj913Eil3OQHMArgZ6p6WERmATgkIvuy2A5Vfbp+3SOienGTX1X7APRlvw+JyIcAFta7Y0RUXxN6fyEi1wP4FoCD2UWbReRdEdkpIuPuwyUinSLSLSLd3ts8ImqcipNfRGYC+B2An6rqGQC/ALAEwHIU3xn8fLx2qtqlqh2q2tGozzJE5KsoG0WkFcXE362qvwcAVT2lqmOqehHALwGsqF83iajW3OSX4teSvwLwoapuL7m8veRqGwC8X/vuEVG9VPJt/20Avg/gPRF5O7vsEQAPishyFMt/PQB+XJcelqjnlF+vHGd9X+F9nPFKUh7vu5KUEqlXTvPKjCnf43hTV+u1jTWQXgL1jrn3nLDuP7XMWKlKvu3/E4DxepNU0yeifPEbOKKgmPxEQTH5iYJi8hMFxeQnCorJTxRUJXX+ppFSz06ZYgnYddvUYcup2z2nbF3uHZfU6ciW1L7lOQXck9L3Ri2nz1d+oqCY/ERBMfmJgmLyEwXF5CcKislPFBSTnyiohm7RLSKnAZwouWgegE8a1oGJada+NWu/APatWrXs21+r6vxKrtjQ5P/KnYt0q2pHbh0wNGvfmrVfAPtWrbz6xrf9REEx+YmCyjv5u3K+f0uz9q1Z+wWwb9XKpW+5fuYnovzk/cpPRDnJJflF5B4R+V8ROSYiW/PoQzki0iMi74nI2yLSnXNfdopIv4i8X3LZ1SKyT0SOZj/H3SYtp75tE5He7Ni9LSJrc+rbIhH5TxE5IiIfiMg/ZJfneuyMfuVy3Br+tl9EWgD8H4DvAjgJ4C0AD6rqkYZ2pAwR6QHQoaq514RF5E4AwwCeV9Ubs8v+GcCAqj6V/cc5V1X/sUn6tg3AcN47N2cbyrSX7iwN4D4AP0COx87o1wPI4bjl8cq/AsAxVT2uqhcA/AbA+hz60fRU9QCAgcsuXg9gV/b7LhSfPA1Xpm9NQVX7VPVw9vsQgEs7S+d67Ix+5SKP5F8I4M8lf59Ec235rQD+KCKHRKQz786MY0G2bToAfAxgQZ6dGYe7c3MjXbazdNMcu2p2vK41fuH3Vber6t8BWAPgJ9nb26akxc9szVSuqWjn5kYZZ2fpv8jz2FW743Wt5ZH8vQAWlfz9jeyypqCqvdnPfgAvofl2Hz51aZPU7Gd/zv35i2bauXm8naXRBMeumXa8ziP53wKwVES+KSJTAHwPwN4c+vEVIjIj+yIGIjIDwGo03+7DewFsyn7fBODlHPvyJc2yc3O5naWR87Fruh2vVbXh/wCsRfEb/48APJpHH8r0azGAd7J/H+TdNwB7UHwbOILidyM/BHANgP0AjgL4DwBXN1Hf/hXAewDeRTHR2nPq2+0ovqV/F8Db2b+1eR87o1+5HDeO8CMKil/4EQXF5CcKislPFBSTnygoJj9RUEx+oqCY/ERBMfmJgvp/3FvX69W6S4sAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "for i in range(16):\n",
    "    plt.imshow(((x + nn.Sigmoid()(residual))[i] / 2).detach().numpy()[0], cmap='gray')\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Attack Linear Model with Fast Gradient Sign Method\n",
    "\n",
    "https://arxiv.org/pdf/1412.6572.pdf"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "class LinearNet(nn.Module):\n",
    "    def __init__(self):\n",
    "        super(LinearNet, self).__init__()\n",
    "        self.clf = nn.Sequential(\n",
    "            nn.Linear(784, 10),\n",
    "            nn.Softmax()\n",
    "        )\n",
    "        \n",
    "    def forward(self, x):\n",
    "        return self.clf(x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/site-packages/torch/nn/modules/container.py:91: UserWarning: Implicit dimension choice for softmax has been deprecated. Change the call to include dim=X as an argument.\n",
      "  input = module(input)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "----------------------------------------------------------------\n",
      "        Layer (type)               Output Shape         Param #\n",
      "================================================================\n",
      "            Linear-1                   [-1, 10]           7,850\n",
      "           Softmax-2                   [-1, 10]               0\n",
      "================================================================\n",
      "Total params: 7,850\n",
      "Trainable params: 7,850\n",
      "Non-trainable params: 0\n",
      "----------------------------------------------------------------\n",
      "-0.8082277254203953\t-0.8972434944152822\n",
      "-0.8942746454238878\t-0.9089972531318669\n",
      "-0.9027950363318131\t-0.914027033233642\n",
      "-0.9079002985795337\t-0.916987309360505\n",
      "-0.9108283610661825\t-0.9202861198425295\n"
     ]
    }
   ],
   "source": [
    "linear_model = LinearNet()\n",
    "\n",
    "summary(linear_model, (784,))\n",
    "\n",
    "optimizer     = optim.Adadelta(linear_model.parameters())\n",
    "loss_function = nn.NLLLoss()\n",
    "\n",
    "for i in range(5):\n",
    "    #training loop\n",
    "    mean_loss = 0.\n",
    "    for j, (x, y) in enumerate(train_loader):\n",
    "        optimizer.zero_grad()\n",
    "        pred = linear_model(x.view([16, 784]))\n",
    "        loss = loss_function(pred, y)\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "        mean_loss = mean_loss + (loss.item() - mean_loss) / float(j + 1)\n",
    "    print(mean_loss, end='\\t')\n",
    "    #testing loop\n",
    "    mean_loss = 0.\n",
    "    for j, (x, y) in enumerate(test_loader):\n",
    "        optimizer.zero_grad()\n",
    "        pred = linear_model(x.view([16, 784]))\n",
    "        loss = loss_function(pred, y)\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "        mean_loss = mean_loss + (loss.item() - mean_loss) / float(j + 1)\n",
    "    print(mean_loss)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/site-packages/torch/nn/modules/container.py:91: UserWarning: Implicit dimension choice for softmax has been deprecated. Change the call to include dim=X as an argument.\n",
      "  input = module(input)\n"
     ]
    }
   ],
   "source": [
    "residual = torch.zeros(x.shape)\n",
    "residual.requires_grad = True\n",
    "\n",
    "aversarial_loss = nn.NLLLoss()\n",
    "adversarial_optimizer = optim.SGD([residual], lr=1000)\n",
    "\n",
    "pred = linear_model((x + residual).view([16, 784])) \n",
    "adversarial_optimizer.zero_grad()\n",
    "adv_loss = -aversarial_loss(pred, y)\n",
    "adv_loss.backward()\n",
    "adversarial_optimizer.step()\n",
    "\n",
    "residual = residual.grad.sign()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/site-packages/torch/nn/modules/container.py:91: UserWarning: Implicit dimension choice for softmax has been deprecated. Change the call to include dim=X as an argument.\n",
      "  input = module(input)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(tensor([ 3,  8,  8,  9,  3,  6,  7,  3,  4,  8,  2,  2,  3,  8,\n",
       "          8,  6]),\n",
       " tensor([ 1,  2,  3,  4,  5,  6,  7,  8,  9,  0,  1,  2,  3,  4,\n",
       "          5,  6]))"
      ]
     },
     "execution_count": 59,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred = linear_model((x - residual * .07).view([16, 784])) \n",
    "pred.argmax(1), y"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(tensor(-1.), tensor(1.))"
      ]
     },
     "execution_count": 51,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "residual.min(), residual.max()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "(0.26785713, 0.9282526)"
      ]
     },
     "execution_count": 60,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "plt.imshow(residual.detach().numpy()[0, 0], cmap='gray')\n",
    "plt.show()\n",
    "\n",
    "residual.detach().numpy()[0, 0].mean(), residual.detach().numpy()[0, 0].var()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow((x - residual * .07).detach().numpy()[1, 0], cmap='gray')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(residual[1][0].detach().numpy(), cmap='gray')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Iterative FGSM"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 227,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "last loss value :  0.011537943035364151\n"
     ]
    }
   ],
   "source": [
    "#FIXME\n",
    "residual = torch.zeros(x.shape)\n",
    "residual.requires_grad = True\n",
    "\n",
    "aversarial_loss = nn.NLLLoss()\n",
    "adversarial_optimizer = optim.SGD([residual], lr=10)\n",
    "\n",
    "x_copy = deepcopy(x)\n",
    "\n",
    "adv_loss_history = []\n",
    "for i in range(100):\n",
    "    #reset the residual noise\n",
    "    residual = torch.zeros(x.shape)\n",
    "    residual.requires_grad = True\n",
    "    #make pred and compute loss\n",
    "    pred = linear_model((x_copy - residual).view([16, 784])) \n",
    "    adversarial_optimizer.zero_grad()\n",
    "    adv_loss = -aversarial_loss(pred, y) #+ 0.00005 * (residual.sign() ** 2).sum()\n",
    "    adv_loss.backward()\n",
    "    adversarial_optimizer.step()\n",
    "    #modify the image\n",
    "    x_copy.data = torch.clamp(x_copy + 0.001 * residual.grad.sign(), 0, 1) \n",
    "    adv_loss_history.append(adv_loss.item())\n",
    "    if not i % 10:\n",
    "        clear_output()\n",
    "        plt.title(\"Adversarial Loss History\")\n",
    "        plt.plot(adv_loss_history)\n",
    "        plt.grid()\n",
    "        plt.show()\n",
    "        print(\"last loss value : \", adv_loss_history[-1])\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 234,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/site-packages/torch/nn/modules/container.py:91: UserWarning: Implicit dimension choice for softmax has been deprecated. Change the call to include dim=X as an argument.\n",
      "  input = module(input)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(tensor([ 3,  8,  8,  4,  5,  6,  7,  3,  4,  0,  1,  2,  3,  8,\n",
       "          8,  6]),\n",
       " tensor([ 1,  2,  3,  4,  5,  6,  7,  8,  9,  0,  1,  2,  3,  4,\n",
       "          5,  6]))"
      ]
     },
     "execution_count": 234,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred = linear_model((x + (x_copy - x) * .5).view([16, 784])) \n",
    "pred.argmax(1), y"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 236,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMi4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvhp/UCwAAEGZJREFUeJzt3WuMXPV5x/Hfs7vcbAPCDbWMcbsUISTLolxWdo2sKlVJ5Fjc/AIIb3AsFAcplhopEkUgKG+4qCoJRlSITW1il5QkUrDAXNpQiCAIHDCYgolboGYRWL4hLPACNuzs0xc7SxfY+Z/Zc+bMOevn+5GsnZ1nzjmPx/PzmTP/Oedv7i4A8fRU3QCAahB+ICjCDwRF+IGgCD8QFOEHgiL8QFCEHwiK8ANB9XVzYz09Pd7Tk///m0aj0cFuvqy3tzf3toss244yeytbmb1lPa+p9Wd9s3V0dDRXT+1sWyr3tezu1s7jCoXfzJZJWiupV9K/uPvtqcf39PRo1qxZubf34Ycf5l42S1ZfqW0XWbYdZfZWtjJ7y3peU+vPCv9HH32Uq6d2ti2V+1puV+7dsJn1SvpnSd+RtEDSlWa2oFONAShXkWP+RZLecved7v6ZpF9KuqQzbQEoW5Hwz5P07oTf32ve9yVmttrMtprZ1qLHUQA6p/QP/Nx9UNKgJPX19XH+MFATRfb8uyTNn/D7qc37AEwDRcL/oqQzzOw0Mzta0nclPdyZtgCULffbfncfMbM1kv5DY0N969399dQyjUajsiGOE088sbTli/6dsnrLWn+Rv1vZvdd122ZtDYXn3nYdhvKyFDrmd/fHJD3WoV4AdBFf7wWCIvxAUIQfCIrwA0ERfiAowg8EZd2csae3t9dnzpzZsn7w4MGu9TJVZY7zI5+i390oU5WviXbP52fPDwRF+IGgCD8QFOEHgiL8QFCEHwiqq5fuNjMVuXR36oqow8PDyWWLnoLJcF79lHmqc9mqPBV6HHt+ICjCDwRF+IGgCD8QFOEHgiL8QFCEHwiqq+P8WY6EyyFHU+a/WZ3H6bNOhS+z99S6s77vMhF7fiAowg8ERfiBoAg/EBThB4Ii/EBQhB8IqtClu81sSNJBSQ1JI+4+kPH4QtcJ5/LZk5s9e3bL2ooVK5LL3njjjYW2nXV9htHR0Za1zZs3J5ddu3Ztsr5///5kPaLh4WGNjIy0denuTnzJ52/c/f0OrAdAF/G2HwiqaPhd0m/N7CUzW92JhgB0R9G3/UvdfZeZ/amkJ8zsv939mYkPaP6nwH8MQM0U2vO7+67mz32SNklaNMljBt19IOvDQADdlTv8ZjbTzI4fvy3p25K2d6oxAOUq8rZ/jqRNZja+nn9z93/vSFcASpc7/O6+U9JfdrCXUtX5WgHnnXdesn7OOeck6zfccEPubRedon3+/Pm5l12zZk2y/v776RHkO++8M/e2pzOu2w+gEMIPBEX4gaAIPxAU4QeCIvxAUIVO6Z2qvr4+T02zXaYqh/IWLfraFx+/ZN26dcn6woULk/WhoaGpttS2rOG2GTNm5K739/cnl3333XeT9SVLliTrU7mM9ZFiKqf0sucHgiL8QFCEHwiK8ANBEX4gKMIPBEX4gaBqNc6fNRZf10t3n3vuucn6fffdl6yfddZZyXqRcfyssfSs3m666aZk/eijj07Wt2zZ0rJ28sknJ5fN8uyzzybrF154YaH111XWa93dGecH0BrhB4Ii/EBQhB8IivADQRF+ICjCDwTViVl629ZoNJJjlFmX165SqrdVq1Yll80ax89yyimnJOsjIyMta8uXL08uu23btmR9z549yXqW+++/v2XtsssuSy576qmnJutFvycwXaVei1O5hgF7fiAowg8ERfiBoAg/EBThB4Ii/EBQhB8IKnOc38zWS7pQ0j53X9i8b7akX0nqlzQk6XJ3P1C0mSLn81f5HYF58+aVuv6nn346WV+9enXL2oEDxf5Zik5tvnPnzpa1RqORq6dxJ5xwQrI+d+7clrXdu3cX2naVRkdHW9amcn2Odvb8P5e07Cv3XSfpSXc/Q9KTzd8BTCOZ4Xf3ZyR98JW7L5G0oXl7g6RLO9wXgJLlPeaf4+7j75v2SJrToX4AdEnh7/a7u5tZywMNM1stqfVBKYBK5N3z7zWzuZLU/Lmv1QPdfdDdB9x9IOe2AJQgb/gflrSyeXulpIc60w6AbskMv5k9IOl5SWea2XtmdrWk2yV9y8zelHRB83cA00hXr9uf+mxAKj6mXGTdRWSN8y9evLjQ+h999NFk/fDhw4XWX6ZHHnmkZW3BggXJZWfPnp2sR71uf8rw8LBGRka4bj+A1gg/EBThB4Ii/EBQhB8IivADQXX10t1Z6jqUl2XXrl3J+oMPPtilTrova4ru008/vWUtaygvdeqqJO3duzdZP1J1ajp69vxAUIQfCIrwA0ERfiAowg8ERfiBoAg/EFStxvmzpMbyi1z2G60dd9xxyfqtt96arKcun50lddlvSbr66qtzr7tqRaaqZ4puAIUQfiAowg8ERfiBoAg/EBThB4Ii/EBQ02qcP6XO4/idOv+6lTL/7kuWLEnWV61alXvdWefrX3vttbnXLU3f60OU/XoZx54fCIrwA0ERfiAowg8ERfiBoAg/EBThB4LKHOc3s/WSLpS0z90XNu+7WdL3Je1vPux6d38sa109PT2aMWNGy/pUzkX+qm6NjdZRkXPDBwYGkvX169fn6mncoUOHWtayrgXw1FNPJetV/psfCa+3dvb8P5e0bJL7f+ruZzf/ZAYfQL1kht/dn5H0QRd6AdBFRY7515jZq2a23sxO6lhHALoib/jvkXS6pLMl7ZZ0R6sHmtlqM9tqZlvdPefmAHRarvC7+153b7j7qKSfSVqUeOyguw+4+4CZ5e0TQIflCr+ZTbwk6wpJ2zvTDoBuaWeo7wFJ35T0DTN7T9I/SPqmmZ0tySUNSfpBiT0CKIF18zjczJIbyxqTnq5jq8cee2yy3tvbm6x//PHHyXrqcCrruvv33ntvsr506dJkvb+/P1kvsuw777yTe92RuXtbx9d8ww8IivADQRF+ICjCDwRF+IGgCD8QVFcv3d3b26tZs2blXr7IFN1VOuaYY0pd/1FHHdWylnVK7hVXXJGsDw0NFaqvXLmyZY2hvGqx5weCIvxAUIQfCIrwA0ERfiAowg8ERfiBoLo6zt9oNApdZnq6+vzzz5P1rFN6sxx//PEta4sXL04umzVOX9SOHTta1rK+/zAyMpKsNxqNXD1hDHt+ICjCDwRF+IGgCD8QFOEHgiL8QFCEHwiqVufzZ52TP12/B/DJJ58UWn7mzJnJ+l133VVo/WV64YUXci+7cePGZP2WW25J1j/77LPc246APT8QFOEHgiL8QFCEHwiK8ANBEX4gKMIPBJU5zm9m8yVtlDRHkksadPe1ZjZb0q8k9UsaknS5ux8o0syROkV3UStWrEjWzz///C510l1XXXVVsp41jn/HHXe0rB06dChXT0eSdvb8I5J+7O4LJP2VpB+a2QJJ10l60t3PkPRk83cA00Rm+N19t7u/3Lx9UNIOSfMkXSJpQ/NhGyRdWlaTADpvSsf8ZtYv6RxJf5A0x913N0t7NHZYAGCaaPu7/WY2S9JvJP3I3T8ysy9q7u5m5i2WWy1pdfN2sW4BdExbe34zO0pjwf+Fuz/YvHuvmc1t1udK2jfZsu4+6O4D7j7Q08PgAlAXmWm0sd31Okk73P0nE0oPSxqfgnWlpIc63x6Aspj7pO/W//8BZksl/V7Sa5JGm3dfr7Hj/l9L+jNJ72hsqO+D1Lr6+vq8rFN6qxwGLDpEecEFFyTr99xzT7Le15f/zOz+/v7cy0rlX/q7iIsuuqhlbfv27V3spLNSr7fh4WGNjIy0dXyd+apx92cltVrZ37azEQD1w0E4EBThB4Ii/EBQhB8IivADQRF+IKiuXrq7qLqe0pvV18KFC5P1u+++O1kvcxz/8ccfT9Y3b96crD///PPJ+mmnndayljrlVuLr4K10Kgfs+YGgCD8QFOEHgiL8QFCEHwiK8ANBEX4gqK6O8zcajeQY5ZF66e6sqaTPPPPMQusfHh5uWbvmmmuSy27atClZP3z4cK6exr399tsta4ODg8llb7vttmR9y5YtyfqBA4WuJF9bWefzt4s9PxAU4QeCIvxAUIQfCIrwA0ERfiAowg8ENa3O50+p8jsCWdvetm1bsn7xxRcn66Ojo8n6smXLWtbqfH36devWJevPPfdcsv7GG28k659++mnLWtHXS9nLdwN7fiAowg8ERfiBoAg/EBThB4Ii/EBQhB8Iytw9/QCz+ZI2SpojySUNuvtaM7tZ0vcl7W8+9Hp3fyy1rr6+Pp81a1bLetZ49sGDB5P1lDK/B1CHMVtMTdnj+EUU3ba7tzXhQTtf8hmR9GN3f9nMjpf0kpk90az91N3/qZ0NAaiXzPC7+25Ju5u3D5rZDknzym4MQLmmdMxvZv2SzpH0h+Zda8zsVTNbb2YntVhmtZltNbOtWW/rAXRP2+E3s1mSfiPpR+7+kaR7JJ0u6WyNvTOYdOI1dx909wF3H+jp4fNFoC7aSqOZHaWx4P/C3R+UJHff6+4Ndx+V9DNJi8prE0CnZYbfxqZKXSdph7v/ZML9cyc8bIWk+p4+BuBr2hnqWyrp95JekzR+0H69pCs19pbfJQ1J+kHzw8HUupIbK3PIrM5DO0Wleq9z30UVeb3U+XmpzVCfuz8rabKVJcf0AdQbn8ABQRF+ICjCDwRF+IGgCD8QFOEHgjpiLt2dpc6n9NZ5zDlLld/NqPJ5K/PS3d36e7HnB4Ii/EBQhB8IivADQRF+ICjCDwRF+IGgMs/n7+jGzPZLemfCXd+Q9H7XGpiauvZW174kesurk739ubuf3M4Duxr+r23cbKu7D1TWQEJde6trXxK95VVVb7ztB4Ii/EBQVYd/sOLtp9S1t7r2JdFbXpX0VukxP4DqVL3nB1CRSsJvZsvM7H/M7C0zu66KHloxsyEze83MXjGzrRX3st7M9pnZ9gn3zTazJ8zszebPSadJq6i3m81sV/O5e8XMllfU23wz+52Z/dHMXjezv2veX+lzl+irkuet62/7zaxX0huSviXpPUkvSrrS3f/Y1UZaMLMhSQPuXvmYsJn9taRhSRvdfWHzvn+U9IG73978j/Mkd//7mvR2s6Thqmdubk4oM3fizNKSLpX0PVX43CX6ulwVPG9V7PkXSXrL3Xe6+2eSfinpkgr6qD13f0bSB1+5+xJJG5q3N2jsxdN1LXqrBXff7e4vN28flDQ+s3Slz12ir0pUEf55kt6d8Pt7qteU3y7pt2b2kpmtrrqZScyZMDPSHklzqmxmEpkzN3fTV2aWrs1zl2fG607jA7+vW+ru50r6jqQfNt/e1pKPHbPVabimrZmbu2WSmaW/UOVzl3fG606rIvy7JM2f8Pupzftqwd13NX/uk7RJ9Zt9eO/4JKnNn/sq7ucLdZq5ebKZpVWD565OM15XEf4XJZ1hZqeZ2dGSvivp4Qr6+Bozm9n8IEZmNlPSt1W/2YcflrSyeXulpIcq7OVL6jJzc6uZpVXxc1e7Ga/dvet/JC3X2Cf+/yvphip6aNHXX0j6r+af16vuTdIDGnsb+LnGPhu5WtKfSHpS0puS/lPS7Br19q8am835VY0FbW5FvS3V2Fv6VyW90vyzvOrnLtFXJc8b3/ADguIDPyAowg8ERfiBoAg/EBThB4Ii/EBQhB8IivADQf0fM9nODIyOGysAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow((x + (x_copy - x) * .5).detach().numpy()[1, 0], cmap='gray')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "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": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}