Adversarial_Attack_On_CIFAR_10.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Adversarial Attack On CIFAR-10.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "toc_visible": true,
      "authorship_tag": "ABX9TyOKNyqqrAN2Ji82RG5PnTxK",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU",
    "widgets": {
      "application/vnd.jupyter.widget-state+json": {
        "6648ff8c7724488ca9183e8ae3447cee": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_44927769419e49e3a2407a051637cc6a",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_9906b04e483b4174aa40bd1590685631",
              "IPY_MODEL_070a3bc8aa8b46a5a42455b7da204410"
            ]
          }
        },
        "44927769419e49e3a2407a051637cc6a": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "9906b04e483b4174aa40bd1590685631": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_4e8d7d4ad78549cdab3fe45b27f94913",
            "_dom_classes": [],
            "description": "",
            "_model_name": "FloatProgressModel",
            "bar_style": "info",
            "max": 1,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 1,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_5ec8ba4584de467da294d149a4a27b52"
          }
        },
        "070a3bc8aa8b46a5a42455b7da204410": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_e3f1aa8b22004830b54e0b0eae660ee6",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 170500096/? [00:30&lt;00:00, 15117284.72it/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_41f583f66e9b485eb7c9b5b5342a62ca"
          }
        },
        "4e8d7d4ad78549cdab3fe45b27f94913": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "5ec8ba4584de467da294d149a4a27b52": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "e3f1aa8b22004830b54e0b0eae660ee6": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "41f583f66e9b485eb7c9b5b5342a62ca": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        }
      }
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/antoninodimaggio/PyTorch-Adversarial-Examples/blob/master/Adversarial_Attack_On_CIFAR_10.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "k1IqEEnf5qEA",
        "colab_type": "text"
      },
      "source": [
        "# Adversarial Attack On CIFAR-10\n",
        "* An adversarial example is when you mess up an image in order to fool a Convolutional Neural Network (CNN) into classifying an image into a different class.\n",
        "* Research into how to prevent adversarial attacks is a very hot topic since the use of CNNs in production becomes more and more common everyday.\n",
        "* This notebook focuses on an attack method known as the HopSkipJumpAttack. This attack can produce images that do not differ in appearance from the original yet will be classified as the wrong label.\n",
        "* First we have to train a fairly robust CIFAR-10 classifier using PyTorch.\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3JHGlQYSwHFP",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import torch\n",
        "import torchvision\n",
        "import torch.nn as nn\n",
        "import torchvision.transforms as transforms"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "67uZLxNjw_vF",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "class RobustCifar(nn.Module):\n",
        "\n",
        "\n",
        "    def __init__(self):\n",
        "        super(RobustCifar, self).__init__()\n",
        "\n",
        "        self.conv_layer = nn.Sequential(  \n",
        "            # block 1\n",
        "            nn.Conv2d(3, 32, 3, padding=1),\n",
        "            nn.ELU(inplace=True),\n",
        "            nn.BatchNorm2d(32),\n",
        "            nn.Conv2d(32, 32, 3, padding=1),\n",
        "            nn.ELU(inplace=True),\n",
        "            nn.BatchNorm2d(32),\n",
        "            nn.MaxPool2d(2),\n",
        "            nn.Dropout2d(p=0.2),\n",
        "            # block 2\n",
        "            nn.Conv2d(32, 64, 3, padding=1),\n",
        "            nn.ELU(inplace=True),\n",
        "            nn.BatchNorm2d(64),\n",
        "            nn.Conv2d(64, 64, 3, padding=1),\n",
        "            nn.ELU(inplace=True),\n",
        "            nn.BatchNorm2d(64),\n",
        "            nn.MaxPool2d(2),\n",
        "            nn.Dropout2d(p=0.3),\n",
        "            # block 3\n",
        "            nn.Conv2d(64, 128, 3, padding=1),\n",
        "            nn.ELU(inplace=True),\n",
        "            nn.BatchNorm2d(128),\n",
        "            nn.Conv2d(128, 128, 3, padding=1),\n",
        "            nn.ELU(inplace=True),\n",
        "            nn.BatchNorm2d(128),\n",
        "            nn.MaxPool2d(2),\n",
        "            nn.Dropout2d(p=0.4)\n",
        "        )\n",
        "\n",
        "        self.fc_layer = nn.Sequential(  \n",
        "            nn.Linear(2048, 10),\n",
        "            nn.Softmax(dim=1)\n",
        "        )\n",
        "\n",
        "\n",
        "    def forward(self, x):\n",
        "        x = self.conv_layer(x)\n",
        "        x = x.view(x.size(0), -1)\n",
        "        x = self.fc_layer(x)\n",
        "        return x\n",
        "\n",
        "\n",
        "def test():\n",
        "    model = RobustCifar()\n",
        "    x = torch.zeros(1,3,32,32)\n",
        "    y = model(x)\n",
        "    print(y)\n",
        "# test()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zdEHJ-JPyXSw",
        "colab_type": "code",
        "outputId": "2130e728-aaca-4c7e-fcde-5f7d8a2fcd1c",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 100,
          "referenced_widgets": [
            "6648ff8c7724488ca9183e8ae3447cee",
            "44927769419e49e3a2407a051637cc6a",
            "9906b04e483b4174aa40bd1590685631",
            "070a3bc8aa8b46a5a42455b7da204410",
            "4e8d7d4ad78549cdab3fe45b27f94913",
            "5ec8ba4584de467da294d149a4a27b52",
            "e3f1aa8b22004830b54e0b0eae660ee6",
            "41f583f66e9b485eb7c9b5b5342a62ca"
          ]
        }
      },
      "source": [
        "# data augmentation for training\n",
        "train_transform = transforms.Compose([\n",
        "    transforms.RandomCrop(32, padding=4),\n",
        "    transforms.RandomHorizontalFlip(),\n",
        "    transforms.ToTensor(),\n",
        "    transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2470, 0.2435, 0.2616))])\n",
        "\n",
        "test_transform = transforms.Compose(\n",
        "    [transforms.ToTensor(),\n",
        "     transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2470, 0.2435, 0.2616))])\n",
        "\n",
        "\n",
        "train_set = torchvision.datasets.CIFAR10(root='./data/downloads', train=True,\n",
        "                                        download=True, transform=train_transform)\n",
        "\n",
        "\n",
        "test_set = torchvision.datasets.CIFAR10(root='./data/downloads', train=False,\n",
        "                                       download=True, transform=test_transform)\n",
        "\n",
        "test_set, validation_set = torch.utils.data.random_split(test_set, [5000, 5000])\n",
        "\n",
        "# 50,000 training samples\n",
        "train_loader = torch.utils.data.DataLoader(train_set, batch_size=64,\n",
        "                                          shuffle=True, num_workers=2,\n",
        "                                          pin_memory=True)\n",
        "\n",
        "# 5,000 validation samples\n",
        "validation_loader = torch.utils.data.DataLoader(test_set, batch_size=64,\n",
        "                                         shuffle=True, num_workers=2,\n",
        "                                         pin_memory=True)\n",
        "\n",
        "# 5,000 test samples\n",
        "test_loader = torch.utils.data.DataLoader(test_set, batch_size=1,\n",
        "                                         shuffle=True, num_workers=2,\n",
        "                                         pin_memory=True)\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Downloading https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz to ./data/downloads/cifar-10-python.tar.gz\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "6648ff8c7724488ca9183e8ae3447cee",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=1.0, bar_style='info', max=1.0), HTML(value='')))"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "Extracting ./data/downloads/cifar-10-python.tar.gz to ./data/downloads\n",
            "Files already downloaded and verified\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DL7-yZeazWnI",
        "colab_type": "code",
        "outputId": "32a598e0-4180-43f8-828e-1e8340bc4ff4",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "# would not want to train this on a cpu\n",
        "device = torch.device('cuda')\n",
        "model = RobustCifar().to(device)\n",
        "criterion = nn.CrossEntropyLoss()\n",
        "\"\"\"\n",
        "# ~ 81% 125 epochs\n",
        "optimizer = torch.optim.Adam(model.parameters(), lr=0.001, weight_decay=0.000001)\n",
        "scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer, milestones=[75, 100], gamma=0.5)\n",
        "\"\"\"\n",
        "# ~ 85% 120 epochs\n",
        "optimizer = torch.optim.Adam(model.parameters(), lr=0.001, weight_decay=0.0001)\n",
        "scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer, milestones=[75, 100], gamma=0.5)\n",
        "\n",
        "\n",
        "num_epochs = 120\n",
        "print('Start training ...')\n",
        "for epoch in range(num_epochs):\n",
        "  training_loss = 0.0\n",
        "  validation_loss = 0.0\n",
        "  model.train()\n",
        "  for batch, (inputs, labels) in enumerate(train_loader, 0):\n",
        "    inputs, labels = inputs.to(device), labels.to(device)\n",
        "    optimizer.zero_grad()\n",
        "    outputs = model(inputs)\n",
        "    loss = criterion(outputs, labels)\n",
        "    loss.backward()\n",
        "    optimizer.step()\n",
        "    training_loss += loss.item()\n",
        "  print('Training --> Epoch: %d, Loss: %1.8f LR: %1.8f' % (epoch+1, training_loss/len(train_loader), optimizer.param_groups[0]['lr']))\n",
        "  scheduler.step()\n",
        "  model.eval()\n",
        "  for batch, (inputs, labels) in enumerate(validation_loader, 0):\n",
        "    inputs, labels = inputs.to(device), labels.to(device)\n",
        "    with torch.no_grad():\n",
        "      outputs = model(inputs)\n",
        "      loss = criterion(outputs, labels)\n",
        "      validation_loss += loss.item()\n",
        "  print('Validation --> Epoch: %d, Loss: %1.8f' % (epoch+1, validation_loss/len(validation_loader)))\n",
        "print('Done training!')"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Start training ...\n",
            "Training --> Epoch: 1, Loss: 2.11311411 LR: 0.00100000\n",
            "Validation --> Epoch: 1, Loss: 2.06976370\n",
            "Training --> Epoch: 2, Loss: 2.05368709 LR: 0.00100000\n",
            "Validation --> Epoch: 2, Loss: 2.01400391\n",
            "Training --> Epoch: 3, Loss: 2.01073951 LR: 0.00100000\n",
            "Validation --> Epoch: 3, Loss: 1.94448510\n",
            "Training --> Epoch: 4, Loss: 1.97341613 LR: 0.00100000\n",
            "Validation --> Epoch: 4, Loss: 1.89129612\n",
            "Training --> Epoch: 5, Loss: 1.94646561 LR: 0.00100000\n",
            "Validation --> Epoch: 5, Loss: 1.86945141\n",
            "Training --> Epoch: 6, Loss: 1.93035220 LR: 0.00100000\n",
            "Validation --> Epoch: 6, Loss: 1.88402562\n",
            "Training --> Epoch: 7, Loss: 1.91747389 LR: 0.00100000\n",
            "Validation --> Epoch: 7, Loss: 1.86067864\n",
            "Training --> Epoch: 8, Loss: 1.90067140 LR: 0.00100000\n",
            "Validation --> Epoch: 8, Loss: 1.85152982\n",
            "Training --> Epoch: 9, Loss: 1.89376871 LR: 0.00100000\n",
            "Validation --> Epoch: 9, Loss: 1.83027994\n",
            "Training --> Epoch: 10, Loss: 1.88376594 LR: 0.00100000\n",
            "Validation --> Epoch: 10, Loss: 1.81756229\n",
            "Training --> Epoch: 11, Loss: 1.86750693 LR: 0.00100000\n",
            "Validation --> Epoch: 11, Loss: 1.81492978\n",
            "Training --> Epoch: 12, Loss: 1.86107072 LR: 0.00100000\n",
            "Validation --> Epoch: 12, Loss: 1.80506593\n",
            "Training --> Epoch: 13, Loss: 1.85487572 LR: 0.00100000\n",
            "Validation --> Epoch: 13, Loss: 1.78699699\n",
            "Training --> Epoch: 14, Loss: 1.85133856 LR: 0.00100000\n",
            "Validation --> Epoch: 14, Loss: 1.77323132\n",
            "Training --> Epoch: 15, Loss: 1.83976362 LR: 0.00100000\n",
            "Validation --> Epoch: 15, Loss: 1.77965763\n",
            "Training --> Epoch: 16, Loss: 1.83489349 LR: 0.00100000\n",
            "Validation --> Epoch: 16, Loss: 1.75807112\n",
            "Training --> Epoch: 17, Loss: 1.83127946 LR: 0.00100000\n",
            "Validation --> Epoch: 17, Loss: 1.75850004\n",
            "Training --> Epoch: 18, Loss: 1.82823639 LR: 0.00100000\n",
            "Validation --> Epoch: 18, Loss: 1.75577869\n",
            "Training --> Epoch: 19, Loss: 1.82072733 LR: 0.00100000\n",
            "Validation --> Epoch: 19, Loss: 1.75240067\n",
            "Training --> Epoch: 20, Loss: 1.81750028 LR: 0.00100000\n",
            "Validation --> Epoch: 20, Loss: 1.74679515\n",
            "Training --> Epoch: 21, Loss: 1.81519883 LR: 0.00100000\n",
            "Validation --> Epoch: 21, Loss: 1.73479788\n",
            "Training --> Epoch: 22, Loss: 1.80555093 LR: 0.00100000\n",
            "Validation --> Epoch: 22, Loss: 1.72831861\n",
            "Training --> Epoch: 23, Loss: 1.79997577 LR: 0.00100000\n",
            "Validation --> Epoch: 23, Loss: 1.74470870\n",
            "Training --> Epoch: 24, Loss: 1.80259952 LR: 0.00100000\n",
            "Validation --> Epoch: 24, Loss: 1.72378104\n",
            "Training --> Epoch: 25, Loss: 1.79570863 LR: 0.00100000\n",
            "Validation --> Epoch: 25, Loss: 1.73147852\n",
            "Training --> Epoch: 26, Loss: 1.79274089 LR: 0.00100000\n",
            "Validation --> Epoch: 26, Loss: 1.71900072\n",
            "Training --> Epoch: 27, Loss: 1.78971490 LR: 0.00100000\n",
            "Validation --> Epoch: 27, Loss: 1.74163932\n",
            "Training --> Epoch: 28, Loss: 1.78709063 LR: 0.00100000\n",
            "Validation --> Epoch: 28, Loss: 1.73312360\n",
            "Training --> Epoch: 29, Loss: 1.78464827 LR: 0.00100000\n",
            "Validation --> Epoch: 29, Loss: 1.72159574\n",
            "Training --> Epoch: 30, Loss: 1.78004775 LR: 0.00100000\n",
            "Validation --> Epoch: 30, Loss: 1.72156276\n",
            "Training --> Epoch: 31, Loss: 1.77860888 LR: 0.00100000\n",
            "Validation --> Epoch: 31, Loss: 1.70473395\n",
            "Training --> Epoch: 32, Loss: 1.77580469 LR: 0.00100000\n",
            "Validation --> Epoch: 32, Loss: 1.71234900\n",
            "Training --> Epoch: 33, Loss: 1.77410538 LR: 0.00100000\n",
            "Validation --> Epoch: 33, Loss: 1.71039826\n",
            "Training --> Epoch: 34, Loss: 1.77199165 LR: 0.00100000\n",
            "Validation --> Epoch: 34, Loss: 1.71525101\n",
            "Training --> Epoch: 35, Loss: 1.77403304 LR: 0.00100000\n",
            "Validation --> Epoch: 35, Loss: 1.70221970\n",
            "Training --> Epoch: 36, Loss: 1.77092840 LR: 0.00100000\n",
            "Validation --> Epoch: 36, Loss: 1.70852347\n",
            "Training --> Epoch: 37, Loss: 1.76945132 LR: 0.00100000\n",
            "Validation --> Epoch: 37, Loss: 1.70466037\n",
            "Training --> Epoch: 38, Loss: 1.76585702 LR: 0.00100000\n",
            "Validation --> Epoch: 38, Loss: 1.70017089\n",
            "Training --> Epoch: 39, Loss: 1.76147153 LR: 0.00100000\n",
            "Validation --> Epoch: 39, Loss: 1.69534596\n",
            "Training --> Epoch: 40, Loss: 1.75733660 LR: 0.00100000\n",
            "Validation --> Epoch: 40, Loss: 1.69153327\n",
            "Training --> Epoch: 41, Loss: 1.75826148 LR: 0.00100000\n",
            "Validation --> Epoch: 41, Loss: 1.70248278\n",
            "Training --> Epoch: 42, Loss: 1.75972104 LR: 0.00100000\n",
            "Validation --> Epoch: 42, Loss: 1.69484152\n",
            "Training --> Epoch: 43, Loss: 1.75355771 LR: 0.00100000\n",
            "Validation --> Epoch: 43, Loss: 1.69382519\n",
            "Training --> Epoch: 44, Loss: 1.75193154 LR: 0.00100000\n",
            "Validation --> Epoch: 44, Loss: 1.69738925\n",
            "Training --> Epoch: 45, Loss: 1.75388273 LR: 0.00100000\n",
            "Validation --> Epoch: 45, Loss: 1.68690091\n",
            "Training --> Epoch: 46, Loss: 1.75216132 LR: 0.00100000\n",
            "Validation --> Epoch: 46, Loss: 1.69133078\n",
            "Training --> Epoch: 47, Loss: 1.74901529 LR: 0.00100000\n",
            "Validation --> Epoch: 47, Loss: 1.69116565\n",
            "Training --> Epoch: 48, Loss: 1.74831794 LR: 0.00100000\n",
            "Validation --> Epoch: 48, Loss: 1.68906722\n",
            "Training --> Epoch: 49, Loss: 1.75071810 LR: 0.00100000\n",
            "Validation --> Epoch: 49, Loss: 1.67810975\n",
            "Training --> Epoch: 50, Loss: 1.74525196 LR: 0.00100000\n",
            "Validation --> Epoch: 50, Loss: 1.69302538\n",
            "Training --> Epoch: 51, Loss: 1.74474987 LR: 0.00100000\n",
            "Validation --> Epoch: 51, Loss: 1.68235236\n",
            "Training --> Epoch: 52, Loss: 1.74230910 LR: 0.00100000\n",
            "Validation --> Epoch: 52, Loss: 1.67569121\n",
            "Training --> Epoch: 53, Loss: 1.73968012 LR: 0.00100000\n",
            "Validation --> Epoch: 53, Loss: 1.68152287\n",
            "Training --> Epoch: 54, Loss: 1.73818882 LR: 0.00100000\n",
            "Validation --> Epoch: 54, Loss: 1.67271121\n",
            "Training --> Epoch: 55, Loss: 1.73628619 LR: 0.00100000\n",
            "Validation --> Epoch: 55, Loss: 1.67655762\n",
            "Training --> Epoch: 56, Loss: 1.73837036 LR: 0.00100000\n",
            "Validation --> Epoch: 56, Loss: 1.67523869\n",
            "Training --> Epoch: 57, Loss: 1.73580445 LR: 0.00100000\n",
            "Validation --> Epoch: 57, Loss: 1.68726726\n",
            "Training --> Epoch: 58, Loss: 1.73418374 LR: 0.00100000\n",
            "Validation --> Epoch: 58, Loss: 1.67472211\n",
            "Training --> Epoch: 59, Loss: 1.72982570 LR: 0.00100000\n",
            "Validation --> Epoch: 59, Loss: 1.68682492\n",
            "Training --> Epoch: 60, Loss: 1.73156156 LR: 0.00100000\n",
            "Validation --> Epoch: 60, Loss: 1.67048122\n",
            "Training --> Epoch: 61, Loss: 1.73316337 LR: 0.00100000\n",
            "Validation --> Epoch: 61, Loss: 1.67785286\n",
            "Training --> Epoch: 62, Loss: 1.72896344 LR: 0.00100000\n",
            "Validation --> Epoch: 62, Loss: 1.67496022\n",
            "Training --> Epoch: 63, Loss: 1.72980679 LR: 0.00100000\n",
            "Validation --> Epoch: 63, Loss: 1.68799782\n",
            "Training --> Epoch: 64, Loss: 1.73102263 LR: 0.00100000\n",
            "Validation --> Epoch: 64, Loss: 1.66553516\n",
            "Training --> Epoch: 65, Loss: 1.72721270 LR: 0.00100000\n",
            "Validation --> Epoch: 65, Loss: 1.67219466\n",
            "Training --> Epoch: 66, Loss: 1.72598111 LR: 0.00100000\n",
            "Validation --> Epoch: 66, Loss: 1.66545573\n",
            "Training --> Epoch: 67, Loss: 1.72684616 LR: 0.00100000\n",
            "Validation --> Epoch: 67, Loss: 1.67384302\n",
            "Training --> Epoch: 68, Loss: 1.72585578 LR: 0.00100000\n",
            "Validation --> Epoch: 68, Loss: 1.66882321\n",
            "Training --> Epoch: 69, Loss: 1.72476450 LR: 0.00100000\n",
            "Validation --> Epoch: 69, Loss: 1.66928211\n",
            "Training --> Epoch: 70, Loss: 1.72373075 LR: 0.00100000\n",
            "Validation --> Epoch: 70, Loss: 1.66795077\n",
            "Training --> Epoch: 71, Loss: 1.72253061 LR: 0.00100000\n",
            "Validation --> Epoch: 71, Loss: 1.66434785\n",
            "Training --> Epoch: 72, Loss: 1.72148925 LR: 0.00100000\n",
            "Validation --> Epoch: 72, Loss: 1.66675020\n",
            "Training --> Epoch: 73, Loss: 1.71976604 LR: 0.00100000\n",
            "Validation --> Epoch: 73, Loss: 1.66367538\n",
            "Training --> Epoch: 74, Loss: 1.71828717 LR: 0.00100000\n",
            "Validation --> Epoch: 74, Loss: 1.66335280\n",
            "Training --> Epoch: 75, Loss: 1.72270129 LR: 0.00100000\n",
            "Validation --> Epoch: 75, Loss: 1.66924935\n",
            "Training --> Epoch: 76, Loss: 1.70339279 LR: 0.00050000\n",
            "Validation --> Epoch: 76, Loss: 1.64812117\n",
            "Training --> Epoch: 77, Loss: 1.69780914 LR: 0.00050000\n",
            "Validation --> Epoch: 77, Loss: 1.64736162\n",
            "Training --> Epoch: 78, Loss: 1.69270754 LR: 0.00050000\n",
            "Validation --> Epoch: 78, Loss: 1.64597409\n",
            "Training --> Epoch: 79, Loss: 1.69310303 LR: 0.00050000\n",
            "Validation --> Epoch: 79, Loss: 1.64018169\n",
            "Training --> Epoch: 80, Loss: 1.68840252 LR: 0.00050000\n",
            "Validation --> Epoch: 80, Loss: 1.64934988\n",
            "Training --> Epoch: 81, Loss: 1.68900846 LR: 0.00050000\n",
            "Validation --> Epoch: 81, Loss: 1.64500921\n",
            "Training --> Epoch: 82, Loss: 1.68580078 LR: 0.00050000\n",
            "Validation --> Epoch: 82, Loss: 1.64108157\n",
            "Training --> Epoch: 83, Loss: 1.68431274 LR: 0.00050000\n",
            "Validation --> Epoch: 83, Loss: 1.64471688\n",
            "Training --> Epoch: 84, Loss: 1.68300122 LR: 0.00050000\n",
            "Validation --> Epoch: 84, Loss: 1.64191650\n",
            "Training --> Epoch: 85, Loss: 1.68265113 LR: 0.00050000\n",
            "Validation --> Epoch: 85, Loss: 1.63594887\n",
            "Training --> Epoch: 86, Loss: 1.68192626 LR: 0.00050000\n",
            "Validation --> Epoch: 86, Loss: 1.63515834\n",
            "Training --> Epoch: 87, Loss: 1.68092458 LR: 0.00050000\n",
            "Validation --> Epoch: 87, Loss: 1.63875112\n",
            "Training --> Epoch: 88, Loss: 1.68106358 LR: 0.00050000\n",
            "Validation --> Epoch: 88, Loss: 1.63954136\n",
            "Training --> Epoch: 89, Loss: 1.67929792 LR: 0.00050000\n",
            "Validation --> Epoch: 89, Loss: 1.64322716\n",
            "Training --> Epoch: 90, Loss: 1.67873304 LR: 0.00050000\n",
            "Validation --> Epoch: 90, Loss: 1.63998631\n",
            "Training --> Epoch: 91, Loss: 1.67820845 LR: 0.00050000\n",
            "Validation --> Epoch: 91, Loss: 1.63109263\n",
            "Training --> Epoch: 92, Loss: 1.67608850 LR: 0.00050000\n",
            "Validation --> Epoch: 92, Loss: 1.64073133\n",
            "Training --> Epoch: 93, Loss: 1.67459232 LR: 0.00050000\n",
            "Validation --> Epoch: 93, Loss: 1.63465465\n",
            "Training --> Epoch: 94, Loss: 1.67491313 LR: 0.00050000\n",
            "Validation --> Epoch: 94, Loss: 1.63218325\n",
            "Training --> Epoch: 95, Loss: 1.67434310 LR: 0.00050000\n",
            "Validation --> Epoch: 95, Loss: 1.63108901\n",
            "Training --> Epoch: 96, Loss: 1.67430265 LR: 0.00050000\n",
            "Validation --> Epoch: 96, Loss: 1.63810029\n",
            "Training --> Epoch: 97, Loss: 1.67480911 LR: 0.00050000\n",
            "Validation --> Epoch: 97, Loss: 1.63493411\n",
            "Training --> Epoch: 98, Loss: 1.67478950 LR: 0.00050000\n",
            "Validation --> Epoch: 98, Loss: 1.63213839\n",
            "Training --> Epoch: 99, Loss: 1.67344904 LR: 0.00050000\n",
            "Validation --> Epoch: 99, Loss: 1.63201473\n",
            "Training --> Epoch: 100, Loss: 1.67190858 LR: 0.00050000\n",
            "Validation --> Epoch: 100, Loss: 1.62928005\n",
            "Training --> Epoch: 101, Loss: 1.66613910 LR: 0.00025000\n",
            "Validation --> Epoch: 101, Loss: 1.62252698\n",
            "Training --> Epoch: 102, Loss: 1.65949150 LR: 0.00025000\n",
            "Validation --> Epoch: 102, Loss: 1.62236000\n",
            "Training --> Epoch: 103, Loss: 1.66036621 LR: 0.00025000\n",
            "Validation --> Epoch: 103, Loss: 1.62568884\n",
            "Training --> Epoch: 104, Loss: 1.65964650 LR: 0.00025000\n",
            "Validation --> Epoch: 104, Loss: 1.62369778\n",
            "Training --> Epoch: 105, Loss: 1.65692616 LR: 0.00025000\n",
            "Validation --> Epoch: 105, Loss: 1.61953378\n",
            "Training --> Epoch: 106, Loss: 1.65517689 LR: 0.00025000\n",
            "Validation --> Epoch: 106, Loss: 1.61753270\n",
            "Training --> Epoch: 107, Loss: 1.65440833 LR: 0.00025000\n",
            "Validation --> Epoch: 107, Loss: 1.62085455\n",
            "Training --> Epoch: 108, Loss: 1.65495752 LR: 0.00025000\n",
            "Validation --> Epoch: 108, Loss: 1.61618299\n",
            "Training --> Epoch: 109, Loss: 1.65163265 LR: 0.00025000\n",
            "Validation --> Epoch: 109, Loss: 1.62353740\n",
            "Training --> Epoch: 110, Loss: 1.65088931 LR: 0.00025000\n",
            "Validation --> Epoch: 110, Loss: 1.62075410\n",
            "Training --> Epoch: 111, Loss: 1.65326176 LR: 0.00025000\n",
            "Validation --> Epoch: 111, Loss: 1.61490378\n",
            "Training --> Epoch: 112, Loss: 1.65301774 LR: 0.00025000\n",
            "Validation --> Epoch: 112, Loss: 1.61935416\n",
            "Training --> Epoch: 113, Loss: 1.65068016 LR: 0.00025000\n",
            "Validation --> Epoch: 113, Loss: 1.62024005\n",
            "Training --> Epoch: 114, Loss: 1.65217674 LR: 0.00025000\n",
            "Validation --> Epoch: 114, Loss: 1.61846362\n",
            "Training --> Epoch: 115, Loss: 1.64991172 LR: 0.00025000\n",
            "Validation --> Epoch: 115, Loss: 1.61755085\n",
            "Training --> Epoch: 116, Loss: 1.64928284 LR: 0.00025000\n",
            "Validation --> Epoch: 116, Loss: 1.61328651\n",
            "Training --> Epoch: 117, Loss: 1.64842673 LR: 0.00025000\n",
            "Validation --> Epoch: 117, Loss: 1.61331154\n",
            "Training --> Epoch: 118, Loss: 1.64834940 LR: 0.00025000\n",
            "Validation --> Epoch: 118, Loss: 1.61435603\n",
            "Training --> Epoch: 119, Loss: 1.64655995 LR: 0.00025000\n",
            "Validation --> Epoch: 119, Loss: 1.61523773\n",
            "Training --> Epoch: 120, Loss: 1.64707647 LR: 0.00025000\n",
            "Validation --> Epoch: 120, Loss: 1.61155355\n",
            "Done training!\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GsWwD-6EM5k4",
        "colab_type": "code",
        "outputId": "7f938baf-5282-411d-c5e7-9196710cb3ed",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "correct = 0\n",
        "total = 0\n",
        "model.eval() # make sure we are in eval mode\n",
        "with torch.no_grad():\n",
        "    for images, labels in test_loader:\n",
        "        images, labels = images.to(device), labels.to(device)\n",
        "        outputs = model(images)\n",
        "        _, predicted = torch.max(outputs.data, 1)\n",
        "        total += labels.size(0)\n",
        "        correct += (predicted == labels).sum().item()\n",
        "\n",
        "print('Accuracy of the network on the 5000 test images: %d %%' % (\n",
        "    100 * correct / total))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Accuracy of the network on the 5000 test images: 84 %\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "PrZBsU5J5YA0",
        "colab_type": "text"
      },
      "source": [
        "## Time to Implement the HopSkipJump Attack on CIFAR-10\n",
        "* I am going to be using [IBM's Adversarial Robustness Toolbox (ART)](https://adversarial-robustness-toolbox.readthedocs.io/en/latest/index.html) to help me out\n",
        "* The [paper](https://arxiv.org/abs/1904.02144) that showcases this method\n",
        "* First let's take a look at what some of the images look like in CIFAR-10"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QQHwnRsTbI9I",
        "colab_type": "code",
        "outputId": "b9f4cc6a-94bd-48b2-9b48-916fe7ebf474",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 411
        }
      },
      "source": [
        "!pip install adversarial-robustness-toolbox\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "from art.attacks import HopSkipJump\n",
        "from art.classifiers import PyTorchClassifier"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Collecting adversarial-robustness-toolbox\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/f7/b5/7c7ef44bd2729140930612b4d10af2dbcfa0ca6c9592251c490100b4753a/adversarial_robustness_toolbox-1.2.0-py3-none-any.whl (486kB)\n",
            "\r\u001b[K     |▊                               | 10kB 22.4MB/s eta 0:00:01\r\u001b[K     |█▍                              | 20kB 27.3MB/s eta 0:00:01\r\u001b[K     |██                              | 30kB 29.1MB/s eta 0:00:01\r\u001b[K     |██▊                             | 40kB 19.6MB/s eta 0:00:01\r\u001b[K     |███▍                            | 51kB 12.7MB/s eta 0:00:01\r\u001b[K     |████                            | 61kB 11.5MB/s eta 0:00:01\r\u001b[K     |████▊                           | 71kB 11.6MB/s eta 0:00:01\r\u001b[K     |█████▍                          | 81kB 12.6MB/s eta 0:00:01\r\u001b[K     |██████                          | 92kB 12.4MB/s eta 0:00:01\r\u001b[K     |██████▊                         | 102kB 11.8MB/s eta 0:00:01\r\u001b[K     |███████▍                        | 112kB 11.8MB/s eta 0:00:01\r\u001b[K     |████████                        | 122kB 11.8MB/s eta 0:00:01\r\u001b[K     |████████▊                       | 133kB 11.8MB/s eta 0:00:01\r\u001b[K     |█████████▍                      | 143kB 11.8MB/s eta 0:00:01\r\u001b[K     |██████████▏                     | 153kB 11.8MB/s eta 0:00:01\r\u001b[K     |██████████▉                     | 163kB 11.8MB/s eta 0:00:01\r\u001b[K     |███████████▌                    | 174kB 11.8MB/s eta 0:00:01\r\u001b[K     |████████████▏                   | 184kB 11.8MB/s eta 0:00:01\r\u001b[K     |████████████▉                   | 194kB 11.8MB/s eta 0:00:01\r\u001b[K     |█████████████▌                  | 204kB 11.8MB/s eta 0:00:01\r\u001b[K     |██████████████▏                 | 215kB 11.8MB/s eta 0:00:01\r\u001b[K     |██████████████▉                 | 225kB 11.8MB/s eta 0:00:01\r\u001b[K     |███████████████▌                | 235kB 11.8MB/s eta 0:00:01\r\u001b[K     |████████████████▏               | 245kB 11.8MB/s eta 0:00:01\r\u001b[K     |████████████████▉               | 256kB 11.8MB/s eta 0:00:01\r\u001b[K     |█████████████████▌              | 266kB 11.8MB/s eta 0:00:01\r\u001b[K     |██████████████████▏             | 276kB 11.8MB/s eta 0:00:01\r\u001b[K     |██████████████████▉             | 286kB 11.8MB/s eta 0:00:01\r\u001b[K     |███████████████████▌            | 296kB 11.8MB/s eta 0:00:01\r\u001b[K     |████████████████████▎           | 307kB 11.8MB/s eta 0:00:01\r\u001b[K     |█████████████████████           | 317kB 11.8MB/s eta 0:00:01\r\u001b[K     |█████████████████████▋          | 327kB 11.8MB/s eta 0:00:01\r\u001b[K     |██████████████████████▎         | 337kB 11.8MB/s eta 0:00:01\r\u001b[K     |███████████████████████         | 348kB 11.8MB/s eta 0:00:01\r\u001b[K     |███████████████████████▋        | 358kB 11.8MB/s eta 0:00:01\r\u001b[K     |████████████████████████▎       | 368kB 11.8MB/s eta 0:00:01\r\u001b[K     |█████████████████████████       | 378kB 11.8MB/s eta 0:00:01\r\u001b[K     |█████████████████████████▋      | 389kB 11.8MB/s eta 0:00:01\r\u001b[K     |██████████████████████████▎     | 399kB 11.8MB/s eta 0:00:01\r\u001b[K     |███████████████████████████     | 409kB 11.8MB/s eta 0:00:01\r\u001b[K     |███████████████████████████▋    | 419kB 11.8MB/s eta 0:00:01\r\u001b[K     |████████████████████████████▎   | 430kB 11.8MB/s eta 0:00:01\r\u001b[K     |█████████████████████████████   | 440kB 11.8MB/s eta 0:00:01\r\u001b[K     |█████████████████████████████▋  | 450kB 11.8MB/s eta 0:00:01\r\u001b[K     |██████████████████████████████▍ | 460kB 11.8MB/s eta 0:00:01\r\u001b[K     |███████████████████████████████ | 471kB 11.8MB/s eta 0:00:01\r\u001b[K     |███████████████████████████████▊| 481kB 11.8MB/s eta 0:00:01\r\u001b[K     |████████████████████████████████| 491kB 11.8MB/s \n",
            "\u001b[?25hRequirement already satisfied: Pillow==7.0.0 in /usr/local/lib/python3.6/dist-packages (from adversarial-robustness-toolbox) (7.0.0)\n",
            "Requirement already satisfied: numpy in /usr/local/lib/python3.6/dist-packages (from adversarial-robustness-toolbox) (1.18.4)\n",
            "Requirement already satisfied: matplotlib in /usr/local/lib/python3.6/dist-packages (from adversarial-robustness-toolbox) (3.2.1)\n",
            "Collecting scikit-learn==0.22.1\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/d1/48/e9fa9e252abcd1447eff6f9257636af31758a6e46fd5ce5d3c879f6907cb/scikit_learn-0.22.1-cp36-cp36m-manylinux1_x86_64.whl (7.0MB)\n",
            "\u001b[K     |████████████████████████████████| 7.1MB 6.1MB/s \n",
            "\u001b[?25hRequirement already satisfied: six in /usr/local/lib/python3.6/dist-packages (from adversarial-robustness-toolbox) (1.12.0)\n",
            "Requirement already satisfied: setuptools in /usr/local/lib/python3.6/dist-packages (from adversarial-robustness-toolbox) (46.3.0)\n",
            "Requirement already satisfied: scipy in /usr/local/lib/python3.6/dist-packages (from adversarial-robustness-toolbox) (1.4.1)\n",
            "Requirement already satisfied: cycler>=0.10 in /usr/local/lib/python3.6/dist-packages (from matplotlib->adversarial-robustness-toolbox) (0.10.0)\n",
            "Requirement already satisfied: pyparsing!=2.0.4,!=2.1.2,!=2.1.6,>=2.0.1 in /usr/local/lib/python3.6/dist-packages (from matplotlib->adversarial-robustness-toolbox) (2.4.7)\n",
            "Requirement already satisfied: kiwisolver>=1.0.1 in /usr/local/lib/python3.6/dist-packages (from matplotlib->adversarial-robustness-toolbox) (1.2.0)\n",
            "Requirement already satisfied: python-dateutil>=2.1 in /usr/local/lib/python3.6/dist-packages (from matplotlib->adversarial-robustness-toolbox) (2.8.1)\n",
            "Requirement already satisfied: joblib>=0.11 in /usr/local/lib/python3.6/dist-packages (from scikit-learn==0.22.1->adversarial-robustness-toolbox) (0.14.1)\n",
            "Installing collected packages: scikit-learn, adversarial-robustness-toolbox\n",
            "  Found existing installation: scikit-learn 0.22.2.post1\n",
            "    Uninstalling scikit-learn-0.22.2.post1:\n",
            "      Successfully uninstalled scikit-learn-0.22.2.post1\n",
            "Successfully installed adversarial-robustness-toolbox-1.2.0 scikit-learn-0.22.1\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0bl47Hcwb_xg",
        "colab_type": "code",
        "outputId": "6f5c9e9d-de1b-49d5-e642-7a8de37174a8",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 298
        }
      },
      "source": [
        "classes = ['plane', 'car', 'bird', 'cat',\n",
        "           'deer', 'dog', 'frog', 'horse', 'ship', 'truck']\n",
        "\n",
        "test_normalize = test_transform.transforms[1]\n",
        "mean = torch.tensor(test_normalize.mean, dtype=torch.float32)\n",
        "std = torch.tensor(test_normalize.std, dtype=torch.float32)\n",
        "inverse_transform = transforms.Normalize((-mean / std).tolist(), (1.0 / std).tolist())\n",
        "\n",
        "\n",
        "def imshow(img_tensor):\n",
        "    img_tensor = inverse_transform(torch.squeeze(img_tensor)).permute(1,2,0)\n",
        "    plt.imshow(img_tensor)\n",
        "    plt.show()\n",
        "\n",
        "\n",
        "dataiter = iter(test_loader)\n",
        "image, label = dataiter.next()\n",
        "print('The shape of the target_image tensor is ' + str(image.shape))\n",
        "plt.title('Image label: ' + str(classes[label.item()]))\n",
        "imshow(image)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "The shape of the target_image tensor is torch.Size([1, 3, 32, 32])\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPsAAAEICAYAAACZA4KlAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjEsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+j8jraAAAgAElEQVR4nO2de5Dc1ZXfP6cf0/OeYUbSaJCEJITAFsYISsY4xjZe2xjjssFVG8dkY3DihK3NksQVJ1XEqazZrd0qb2oxcWU3domYNbvGsKxt1tSGjWFZJ5RfgDBYPARI4iWJQaPXzPQ8evp18kf/VGnBPXdGmpkewe98qqZm5p6+v9/p2336132/fc4RVcVxnLc/meV2wHGc1uDB7jgpwYPdcVKCB7vjpAQPdsdJCR7sjpMSPNgdRORyEdk/z9t+QUR+eornOeW5zsLxYF9iRORlEfnocvuRFkRkg4ioiOSW25fTDQ92x0kJHuwtJHkb+zMRuVVExkTkRRH5R8n4PhEZFZHrm27/SRF5QkQmEvvNbzjedSLyiogcEZH/0vwuQkQyInKTiOxN7PeIyMA8/Tw+rygiz4rIZ958E/lTERkXkedE5CNNhj4R+baIjIjIARH5QxHJnuJ6XSYiP0/Wap+IfGEe6/Jw8ntMRCZF5H2ncu63Ix7sree9wE5gEPgecDfwHuAc4J8Bfyoi3cltp4DrgH7gk8DviMg1ACKyBfgfwG8Bw0AfsKbpPP8GuAb4EHAmcAz4s3n6uBf4QHLM3we+KyLDb7gPe4EVwFeBHza9kHwHqCb35yLgCuBfhk4iIn8rIjcZtvXA3wH/HVgJbAWeTMzmugAfTH73q2q3qv5invf57Y+q+s8S/gAvAx9N/v4CsLvJdgGgwFDT2BFgq3Gs/wbcmvz9e8BdTbZOoNx0rl3AR5rsw0AFyAWOezmwP3IfngSubroPrwHSZH8U+DwwBMwCHU22a4GfNM396TzX7T8B987zts3rsiFZ0zfdz7T/+CZG6znY9PcMgKq+cawbQETeC3wNeBfQBhSAv05udyaw7/gkVZ0WkSNNx1kP3Csi9aaxGo2APBBzUESuA/49jcAh8WdF000OaBJZCa8k/qwH8sCIiBy3ZZr9PAnW0Xj3EPIvti6Ogb+NP735HnAfsE5V+4BvAcejaARYe/yGItJB46PBcfYBn1DV/qafdlWdK9DXA7cBNwKDqtoPPN10XoA10hTNwFk0rvb7aFzZVzSds1dVzz/5u84+YJNhi62Lp3EaeLCf3vQAR1W1JCKXAP+0yfZ94FPJBl8bcDMnBuS3gD9KghcRWSkiV8/jnF00AuZQMu+f07iCNrMK+LcikheRfwy8E7hfVUeAB4BbRKQ32STcJCIfOsn7DXAn8FER+ayI5ERkUES2JrbYuhwC6sDZp3DOtzUe7Kc3/xr4AxEp0viMfs9xg6o+Q2MT7m4aV/lJYJTGlRXgGzSufg8k839JY2Mtiqo+C9wC/ILGR44LgJ+94WaPAJuBw8AfAb+pqsc/QlxH4631szQ2Bb9PY7/gTYjI34nIVww/XgWuAr4MHKWxb3BhYo6ty3Ti08+SXfxL57rPaUFO/OjlvFVJdvDHgM2q+tJy++OcfviV/S2MiHxKRDpFpAv4E+ApGrv/jvMmPNjf2lxNY2PsNRpvqz+n/lbNMfC38Y6TEvzK7jgpoaVfqunoaNe+3q5FO14mE3mtirxhyZwgEc97Gta7IMlEjhd55yQRP4SYj6d2zFM5FxH/Y36YSxI5VdQPsR/rTNb+6n1Xd+/JuoFq3bRNjI+btnrdnncqj0vs2Wg9LMfGJpiangmebEHBLiJX0pB4ssD/VNWvxW7f19vFb/2TK8PHigSuFZyFQsGck1V7cdtzedNWjTxg5Vo1ON7WbvtRNeYA5CN+ZCPrUavVTFsuHz5m7AUur3awaNX2v1ovm7a2Qth/ydpP4FwkaHO5DvtcPYOm7dLLPhwcz2bsc1VLM6bt73/8v0zb9MyUaWszHhewLwixFw+thm3fuO0uc84pv41PMpn+DPgEsAW4NknOcBznNGQhn9kvAfao6ouqWqbx5Y75fEPLcZxlYCHBvoYTExz2c2KKJQAicoOI7BCRHdMzpQWcznGchbDku/Gqul1Vt6nqts6O9qU+neM4BgsJ9gM00hCPs5Y5Uicdx1k+FrIb/xiwWUQ20gjyz3Fi9tGbUI1IUZGdRyK7tCYRqaZStXezK5Gd7ryx657NRnZaI35IxFaLrEc9og9WjfvW1tZmHy8iHdaw175es58+VWNeLlIHshZ5CuTF9r9WtRdkfHwsON7Vae/u93Ta54ppdjHFI6agWLJcTLatq3W8iNphWuZAVasiciPwYxrS2+1JJpbjOKchC9LZVfV+4P5F8sVxnCXEvy7rOCnBg91xUoIHu+OkBA92x0kJLc16E7Ez1WIZbJYtliiQz9l3rVK2ZZBMZJ6VrDNbqUSOF5GuYllNpgUykaSKmZlwEkcp4mNsHbMR2VPUfsxytbCc1JOzk4ZiftRr9rk0Ir215cPn6+8PZ8MBFMcPmTYk4uMprmM1IsuZnEJaoV/ZHScleLA7TkrwYHeclODB7jgpwYPdcVJCS3fjVSM7lpEkgliJJotqJKtCIzuW+TZ7t7hm7ZFHElrqkayV6ZKd318p27vnsdJZU5OTwfFYUkW+3U78qFTsslQSKf3V09EdHK/M2r4P9Nk75Nls5Kka2em2FI+ZSOmpajWirkQe6/wplJ6KEUusEaOsVqzUnV/ZHScleLA7TkrwYHeclODB7jgpwYPdcVKCB7vjpISWSm+gpgQRky1inUIsKmVbMioUOk3bbERqqhiS1+zsrDmnOBWWwgBKRtIKwFTRnlePtCeanQ13aZkt291byEXkpEhiUCZSM66/uyc4XuntN+dMF4+attWrVpi2Qt+A7YjVbaUWaV2VsZ+LsefpbMl+HsQSvawVjibIGJJuTOHzK7vjpAQPdsdJCR7sjpMSPNgdJyV4sDtOSvBgd5yU0GLpLUJEM7Da40Rr0EWy1zJiS3m1ui13WC4WJ6fMOUfHwu2HAEYPHjRtI6+/btpqVVsezBjZULEaaJm8/TRoi9TQq0akptJAX3B85tioOeeMSF24em3atA1HWlsVCmFbOVKHcOKoLXvOztoZcdbzFOLtnyw5OpbtWbPk10jW24KCXUReBopADaiq6raFHM9xnKVjMa7sH1bVw4twHMdxlhD/zO44KWGhwa7AAyLyuIjcELqBiNwgIjtEZMfMjP0Zz3GcpWWhb+MvU9UDIrIKeFBEnlPVh5tvoKrbge0AQ6sGTr42j+M4i8KCruyqeiD5PQrcC1yyGE45jrP4nPKVXUS6gIyqFpO/rwD+YI45ZI0sqkqkPVHOKDYYyySSiAZRj8h8mYhENVUsBsdfi8hk48Vx01aKFJwsFNpNW1nsDLaOzvC8oZVD5pyxifD9ApgYO2bahlcMmrZzNqwJjg90d5hzcjn7MatGZK3DR2w5z2qH1d1ly3w9vXYWXayAaDmSWRh7rlqyXEzKU+s6HXnvvJC38UPAvYlDOeB7qvq/F3A8x3GWkFMOdlV9EbhwEX1xHGcJcenNcVKCB7vjpAQPdsdJCR7sjpMSWpz1JmTq4deXtlgvLyPJyzhUg5ydCSUZ2zYZKXq479VXguNHj9ipAbWIzNfbF84MA+jotCWqyaKdZbd65ergeH//GeacY+N2Zt6KlbYMteW8c03bzNiR4PjA6lXmnHe9+wLTVopktu15ecS0YWQ4ZiKFI8sle30lpm1F5F6JPb+NApeVyKm0GpbrYn0M/cruOCnBg91xUoIHu+OkBA92x0kJHuyOkxJauxuvIBreLcxhJ6BYO4z16O6nvdtaKtlJN4dHD5m2scNh28Z1w+ac0SN2IsmKVXYiydSkXXOtOG7XSCsYtfcOvm4nixQju/u9a8K7+xBPCim0hVtsZcSuDdjVY+/8r1qz1rR19tpJPrOl8Dp2DdtzilN2vbg6dt3DWJuyStWeVzbWcdpo5QXQbqgTGlEL/MruOCnBg91xUoIHu+OkBA92x0kJHuyOkxI82B0nJbS8/ZMlDMTSC3JG3bpZtSWSaiSZYfKYnfhx6MCLpq23PSytDPba9czKRsICQHdHWJ4CKEfkQatdEMDBQ2GJbWx8wpxTr9sS5rGjdg29iYGVpm3L5nCSzJr1Z5lz2nrsZJ3+M+xzzRTDdeYAJo0kpdKKfnNOd5dd/88W0OK1DetV+/GcLoYfm8lJW2LtGT4zOB5L1PEru+OkBA92x0kJHuyOkxI82B0nJXiwO05K8GB3nJTQcunNkifKVaPQHGApQ5l2O7OteMyWjA6/9pJpG+i0j7l2bTi7bWzKlrVqJTtzqTRjt38qGW2LIF6fbto4ZrVmi0Yi9mv+4cP2Oh4bsn284D2XBsfXn7XOnNMekSILhS7TNrzSluX+5v88EByfnrSzG99x8VbTNjllZyOWI49ZadqWgmcmwlJwd6edIZjTcLxIRMOe88ouIreLyKiIPN00NiAiD4rI7uS3LZA6jnNaMJ+38d8BrnzD2E3AQ6q6GXgo+d9xnNOYOYM96bf+xvrKVwN3JH/fAVyzyH45jrPInOoG3ZCqHi/W/TqNjq5BROQGEdkhIjusz5OO4yw9C96N18YXtc1tAVXdrqrbVHVbZ4f9nWPHcZaWUw32gyIyDJD8tgucOY5zWnCq0tt9wPXA15LfP5rXLIFMNvz6ks3YxfosuS4TkRkmjPZDACv6bOnqvPWbTFu78c7k0ad2m3OKxaJpyxbsdzoz07aMU63YGVQzJUN6q0cyBMv28Wr2NI5GCl/WjJZG+R4724xIi6SYdDjQa8tyQ73hx3rnY7+w/SjYfux77YBpq0cKRErdlpbPHAivSX+3/Ty13kwvKOtNRO4CfgGcJyL7ReSLNIL8YyKyG/ho8r/jOKcxc17ZVfVaw/SRRfbFcZwlxL8u6zgpwYPdcVKCB7vjpAQPdsdJCadN1lstUqyPTDjtbbZsfyNv9PDrpu2DW883bRvX2b3NxsbDMlpbmy2hzdqKC31d3aZtOpJdNTVt26q18Alj66uRl3yrNx/A4SNv/Bb1/+fnjzwWHN98nr32Kva52mxllrFRWw7buDrcP258LFyIEuD//sPDpq04aUupMQlzeNDuYzcwuCI4Phs5V96QsIn0P/Qru+OkBA92x0kJHuyOkxI82B0nJXiwO05K8GB3nJTQUulNgXq0q5uBIcmUZmfNKaOH7YKCGslA6mhvM23FybD+U6rYqWHZvF1EsR7J5OrutDO5jk7YBS4t5SXbZt8viRT7nK3aa1yctItR/sODDwXHP/yBy80555x3nmnr7rG1t9f32BmOJUNiq0zaBSBf2v2KfTxDBgagbme9TY3Z/QXHDh8Ljq81+rkBDLSHQ1ddenMcx4PdcVKCB7vjpAQPdsdJCR7sjpMSWrsbr2q2eSoU7LZL2Vz4Nak6be+Cd3XZu9k9fb2mLZOxX//KRo2x4pS9Yy05u45Yb59djy3m455X7d3iWi2sdnR2262EyvYmMrOzdrKRVmxlpVwKJ+uUI7X1KiXbdrBk7/yXp+2d7j6jntzYqK3WHDoY3h0HKOXt51wukqxDJCGqUOgJjmvWfu5UseLFd+MdJ/V4sDtOSvBgd5yU4MHuOCnBg91xUoIHu+OkhNZKb/U6JaOTazZS6yxXD8sMHe22NHHuuXZSRXunXfttsmRrJPVsWL7K5m1Zq0NsPebctXa9u3KkZtwT7XbNu8NGnTyJJHDkDWmzMS+SuCSxpKawRNUfkRT7+/pM28iIXVOwOGNrhxmjwN54pFVTod2WgTsiLbu0bteg23DuZtO2ctWa4Ph05LnYZVynY4/IfNo/3S4ioyLydNPYzSJyQESeTH6umus4juMsL/N5G/8d4MrA+K2qujX5uX9x3XIcZ7GZM9hV9WHArhnsOM5bgoVs0N0oIjuTt/lnWDcSkRtEZIeI7Jgp2V8rdRxnaTnVYP8msAnYCowAt1g3VNXtqrpNVbd1tNsbWY7jLC2nFOyqelBVa6paB24DLllctxzHWWxOSXoTkWFVHUn+/QzwdOz2x6nV60xNhrOhZqbt7KpMNixfDQxEWur0m58sOGjU/AKYjrRyGhxaGxzXzB5zTl5sOWZdr10XriSRLMB6pOadIbHVa/Yc1K6FJ3VbzMlGMgQr1fD5itN27bdSJCNu3dBZpu3FSL3BBx55Ijj+xKv7zDmVjL3271ljS7orV640bZleW1b85ZNPBcc7u2Mto8K2mPQ2Z7CLyF3A5cAKEdkPfBW4XES2Jsd+GfjtuY7jOM7yMmewq+q1geFvL4EvjuMsIf51WcdJCR7sjpMSPNgdJyV4sDtOSmhp1psg5IyqfJEkLzJG+6f2gv0lnfZ2+3XswP6XTVt92v6W3wdWhaW3Qt5exqmpSdOWy9sST3ukbVQl0jYqkwn7Uo/JdZEihYVIltdUpNBmW1v4salF/OjpCxdeBKAalmwBqhGZ8pe/2hkc3z9qfwO80BXJzBsYNG3rN5xt2h5/4QXTtu+114Ljw8P280ojcqmFX9kdJyV4sDtOSvBgd5yU4MHuOCnBg91xUoIHu+OkhJZKb9lshu7usLwSK4iYkfBrUqwv26rVZ5q2iaN2ltQrr+w3bTNGxlZPl134cmzMlnjq2UhmW7stvUnezpaTTDhzrFa2pZq2dvt4hYL9uHR02IU7BwfCWYeRh4yjR49E/Ig8VfP2+ncPDAXH9ZWw3AUwEZEU9x0cNW2r19mZeRq548XJ8PPqjJKdCVo2GvTFJDm/sjtOSvBgd5yU4MHuOCnBg91xUoIHu+OkhJbuxtfrdaZLdp0xC3PXPbKDv/7sTaatO1LltlK0Ey4mJyaC4z2ddrJIV8RWEXv5y7N2659S2U4mKRgtsWpV+3i1qp2FtHKl3aLqE1eGeoc06OgI+zE0tMqc091t76pbNe0A1kYSUK77F+GKadU//3Nzzo4dj5u2R416cQArzwwnSgGcFXk+zpZ/HByvVe36hVkjOYxIUpNf2R0nJXiwO05K8GB3nJTgwe44KcGD3XFSgge746SE+XSEWQf8BTBEowPMdlX9hogMAH8FbKDRFeazqmr3VQIkk6HQGZZXNFKErmq0LjoyZp/utUjCwjkbzzFt46N2kszYwZHgeEebndDSE5GTOiP1zKozttTUP7DCtE2+HvZRI+2fCpGkmwvOv8C0ffrTV5u2ru6u4Pj0lN3+aaIYljahkURl0dvXb9q2vfd94eNF2nzt2fsfTdvEIft5le201/H5F18ybXXjud/TY9fkq1YMWS4SR/O5sleBL6vqFuBS4HdFZAtwE/CQqm4GHkr+dxznNGXOYFfVEVX9VfJ3EdgFrAGuBu5IbnYHcM1SOek4zsI5qc/sIrIBuAh4BBhq6uT6Oo23+Y7jnKbMO9hFpBv4AfAlVT3hw5U2PnAHPyyIyA0iskNEdpRKdlEAx3GWlnkFu4jkaQT6nar6w2T4oIgMJ/ZhILhzoarbVXWbqm5rj3wn3XGcpWXOYBcRodGieZeqfr3JdB9wffL39cCPFt89x3EWi/lkvb0f+DzwlIg8mYx9BfgacI+IfBF4BfjsXAdSoGJIQLlIjS6r3tbro7YM8tzuPabtvHPOM23rInXE2mpG5pjR0grg4GFbyqtl7NpvazbYNfQ+duUnTdt37/xu2BBZ34suvMi0ffrT9r6rVRsQoDwb/shWqdgf5QRbHhw7ZstylUhGX74QlgD7InJdlyEPAxyJSFvP7N5r2gZW2XLp4MqVwfEVEWl2xsgerUdq0M0Z7Kr6U+y8uY/MNd9xnNMD/wad46QED3bHSQke7I6TEjzYHScleLA7TkpoacFJVNF6WBo4Gsl4Onok3EJpKpJBNXrIbrsUa8XT29tn2mR1+BvBtYhkNDhoSzy79thSzYc2vsO0XbztPaatvRCWjTLY8tTQKruo5OBgWBYCmDXkNQCphCUqVXut9u593rS9sOtp03bhtvebtk2bzw+Od7TbWW/ZiLxWjmQPPvrkk6Zt/cYNpm3N2nChylhxzuJYuFWWFV/gV3bHSQ0e7I6TEjzYHScleLA7TkrwYHeclODB7jgpoaXSW6VaZfTI4aCtOFE0501Ohm2ZSK+36UihjHK0t5ktd9Smx4Pj9YictPXCd5u2PaMR6fBweJ0AenptOe/jH78iON4b6Tk3Pm6vfcXK9ANE7HU8dizs/3PPP2PO2bt7l2mbHrel1EzeLvQ4fOa64Hh7m/3U37RhvWnbtdeWS2PFWZ5//gXT9qkrPh4cH+i3H+cJY33tR8Sv7I6TGjzYHScleLA7TkrwYHeclODB7jgpocWJMICxE57P2i2U+rrDu5K1yJf+pW7v1Feq9rxyJEGidzi8s1sq2m2o+obsneLMgL13aikQAL09vabtwGv7guOzgwO2H5HEoGNjdg29et3eqV8xEE4oevWlF80540fHTFsuYz9VJ47aO/VT4+GEEYz2VACXfOBS0/b40ztN2+hoWK0ByIv9vHrXO8I1AAfPsH188WVbFbDwK7vjpAQPdsdJCR7sjpMSPNgdJyV4sDtOSvBgd5yUMKf0JiLrgL+g0ZJZge2q+g0RuRn4V8BxbeYrqnp/7FgdHR1s2fKuk3Yynwu7qZFaYbk2u4XP6Mhr9rlW2zXX+oxkkoxR9w2gLvYSr1tvt3gqTk6bto5Ou0FmNhuWHOuRtapUwu21AJ7fvdu0ZSKJMOvP+lBwvKvLlpP2lUqmLR+R3gYibaj6erqD4+WIbLj53HNM21WfDCcaAdTK9nq0F3pM2/lbwud77PGfm3NmKuEkqgW1fwKqwJdV9Vci0gM8LiIPJrZbVfVP5nEMx3GWmfn0ehsBRpK/iyKyC1iz1I45jrO4nNRndhHZAFwEPJIM3SgiO0XkdhE5Y5F9cxxnEZl3sItIN/AD4EuqOgF8E9gEbKVx5b/FmHeDiOwQkR3T0+E2s47jLD3zCnYRydMI9DtV9YcAqnpQVWuqWgduAy4JzVXV7aq6TVW3dUb6XjuOs7TMGewiIsC3gV2q+vWm8eGmm30GsFt2OI6z7MxnN/79wOeBp0TkeH+brwDXishWGnLcy8Bvz3WgfD7HmavDNd7yOTvrDQnLSRljHCCXt49XnrFbTY2O2DJUqTss46xatcKck4lk81G3pZqhVbYEWK3YstGs0Z6oo9Bmzsnkbdv6s20ZqjhuZ/vNzIbX8Z3nX2DOGZ+wHxep2ZJSf6RW22ylEhzv6Q0/lgDj47a02dtrvzvduH6Dbdu42bTtNuTNw0f2m3P6esMycDZrX7/nsxv/UyAUVVFN3XGc0wv/Bp3jpAQPdsdJCR7sjpMSPNgdJyV4sDtOSmhpwclCW4EN6+3WOhY5I+utUrbb7RjJXwB0RVohxYpR1iUbHB+bsv3o6AzPAejpsJ3c/+qrpk0jmU2Txcng+JFDo+acdZtsWWjNurNsP4ZXm7Z2IzNvY+Rczz/3nGmrlexvX8akt/a2sKwYa9UUy6bMRrLvDh6017hUCkuAANPT4QzHLe8835xj9Xlqb/+ZOcWv7I6TEjzYHScleLA7TkrwYHeclODB7jgpwYPdcVJCS6W3XC7HysHBoC1vSCQxKkZGE0COcPYXQFeHndVUqtjz6tnwclXr9jLmIhllsxHpELHltc6I//lc+PW7rWDPyRlzAKqRYpS1iu3/+Fi4/1p5xi6kuensjaYtG5HD2iLPnUo1/Byp1u3HuVq1swoHB+wiofmcvcaRBE1TCpZIIc18Lnyfs5EsS7+yO05K8GB3nJTgwe44KcGD3XFSgge746QED3bHSQktld5EoC0XzgITbKnJkkJimW0dBTuzLVaULx/RSMq18LxCpJhjpE6imc0Hdo8ygJ4eu2/YzEw4OyybtbPvZmfCfcMAzhgIS6UAWrX910rYj4Fu+3hn9HSato6IvFYu2xJscSp836Zn7b5ynZ12P7oL3x2smA7A5EQ44xCgHpGC2/KGpGsrgKbE1tZmy39+ZXeclODB7jgpwYPdcVKCB7vjpAQPdsdJCXPuxotIO/AwUEhu/31V/aqIbATuBgaBx4HPq6qdNQFUKmVGRsItbdpju+fGDn6sZVRsd79Si7Wasl//MsYufnekpl3JaIMEUI0kkrS32T7WIglA1rxKrGXUTNG0jc7YO8xdkUadOQ37WJmx16MjkqwTS07JRJSGru7wznrG2AEH6Oi0VYGs2ve5I5KglMnY/peNx6YtZ59rajr83MlkbDVpPlf2WeA3VPVCGu2ZrxSRS4E/Bm5V1XOAY8AX53Esx3GWiTmDXRscf3nPJz8K/Abw/WT8DuCaJfHQcZxFYb792bNJB9dR4EFgLzCmqsfff+wH1iyNi47jLAbzCnZVranqVmAtcAnwjvmeQERuEJEdIrJjIvINI8dxlpaT2o1X1THgJ8D7gH4ROb7LsRY4YMzZrqrbVHVbb6QntuM4S8ucwS4iK0WkP/m7A/gYsItG0P9mcrPrgR8tlZOO4yyc+STCDAN3iEiWxovDPar6tyLyLHC3iPwh8ATw7bkONDM9w1M7dwZtsTpi+XxYTorVrSu02/JJeyTRYdXKlaZNjBp0E+Pj5pxYskvM/9mSnagRS2oxEyEiSUPdEcloYnzMtJXqtnTY1RGWI8uRtkttxhyAWsR/q84cwKwhU8YkqnKkNmCs7l5Hh/1Yz8za8ma1Gk6S0bp9v2r1sIQZaw02Z7Cr6k7gosD4izQ+vzuO8xbAv0HnOCnBg91xUoIHu+OkBA92x0kJHuyOkxJEI211Fv1kIoeAV5J/VwCHW3ZyG/fjRNyPE3mr+bFeVYP6cUuD/YQTi+xQ1W3LcnL3w/1IoR/+Nt5xUoIHu+OkhOUM9u3LeO5m3I8TcT9O5G3jx7J9Znccp7X423jHSQke7I6TEpYl2EXkShF5XkT2iMhNy+FD4sfLIvKUiDwpIjtaeN7bRWRURJ5uGhsQkQdFZHfy+4xl8uNmETmQrMmTInJVC/xYJyI/EZFnReQZEfl3yXhL1yTiR0vXRETaReRREfl14sfvJ+MbReSRJG7+SkTsHOkQqtrSHyBLo4bd2UAb8GtgS6v9SGMqAPQAAAI1SURBVHx5GVixDOf9IHAx8HTT2H8Fbkr+vgn442Xy42bgP7R4PYaBi5O/e4AXgC2tXpOIHy1dExrVB7qTv/PAI8ClwD3A55LxbwG/czLHXY4r+yXAHlV9URt15u8Grl4GP5YNVX0YOPqG4atpVOmFFlXrNfxoOao6oqq/Sv4u0qiEtIYWr0nEj5aiDRa9ovNyBPsaYF/T/8tZmVaBB0TkcRG5YZl8OM6Qqo4kf78ODC2jLzeKyM7kbf6Sf5xoRkQ20CiW8gjLuCZv8ANavCZLUdE57Rt0l6nqxcAngN8VkQ8ut0PQeGWn8UK0HHwT2ESjIcgIcEurTiwi3cAPgC+p6kSzrZVrEvCj5WuiC6jobLEcwX4AWNf0v1mZdqlR1QPJ71HgXpa3zNZBERkGSH6PLocTqnoweaLVgdto0ZqISJ5GgN2pqj9Mhlu+JiE/lmtNknOfdEVni+UI9seAzcnOYhvwOeC+VjshIl0i0nP8b+AK4On4rCXlPhpVemEZq/UeD66Ez9CCNRERoVGwdJeqfr3J1NI1sfxo9ZosWUXnVu0wvmG38SoaO517gf+8TD6cTUMJ+DXwTCv9AO6i8XawQuOz1xdpNMh8CNgN/D0wsEx+/CXwFLCTRrANt8CPy2i8Rd8JPJn8XNXqNYn40dI1Ad5No2LzThovLL/X9Jx9FNgD/DVQOJnj+tdlHSclpH2DznFSgwe746QED3bHSQke7I6TEjzYHScleLA7TkrwYHeclPD/APgShKGa5h+RAAAAAElFTkSuQmCC\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Bps0fmieq-J7",
        "colab_type": "text"
      },
      "source": [
        "## The Basics of the HopSkipJump attack\n",
        "* This is a powerful black-box attack that only requires final class prediction, and is an advanced version of the boundary attack.\n",
        "In the untargeted setting (target=False) an adversarial is any image for which the predicted label is different from the label of the original image (since HopSkipJump has the classifier we can use it to get the class of the image). Then via various mathematical techniques we can generate an image that will classify into the randomly chosen target class.\n",
        "* NOTE: The model (classifier) can still classify images into the wrong label. When this occurs we can have issues with the adversarial image actually matching the correct label which is something we do not want. This is why targeted attacks (where we essentially pick the class that we are trying to match using an init image) can perform better under certain circumstances. But for the most part this attack will serve its purpose."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "K-QiYsY1d_5s",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "classifier = PyTorchClassifier(model, criterion, optimizer, (3, 32, 32), 10, device_type='gpu')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zdTsufLRgseo",
        "colab_type": "code",
        "outputId": "63c7e660-ea49-421b-9963-253deb81f514",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "attack = HopSkipJump(classifier=classifier, targeted=False, max_iter=0, max_eval=1000, init_eval=10)\n",
        "target_image = image.numpy()\n",
        "print('The label for the image is a ' + classes[label.item()] + '.')\n",
        "classified_label = np.argmax(classifier.predict(target_image))\n",
        "print('The model orginally classfied this image as a ' + classes[classified_label] + '.')\n",
        "\n",
        "iter_step = 10\n",
        "x_adv = None\n",
        "for i in range(20):\n",
        "    x_adv = attack.generate(x=target_image, x_adv_init=x_adv, resume=True)\n",
        "    if(i % 5 == 0):\n",
        "      print(\"Adversarial image at step %d.\" % (i * iter_step), \"L2 error\", \n",
        "          np.linalg.norm(np.reshape(x_adv - target_image, [-1])),\n",
        "          \". Classified as a %s.\" % classes[np.argmax(classifier.predict(x_adv))])\n",
        "      imshow(torch.tensor(x_adv))\n",
        "      plt.show(block=False)\n",
        "    \n",
        "    attack.max_iter = iter_step"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "The label for the image is a cat.\n",
            "The model orginally classfied this image as a dog.\n",
            "Adversarial image at step 0. L2 error 0.8701033 . Classified as a cat.\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "text": [
            "Adversarial image at step 50. L2 error 0.016307417 . Classified as a cat.\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "text": [
            "Adversarial image at step 100. L2 error 0.012010048 . Classified as a cat.\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "stream",
          "text": [
            "Adversarial image at step 150. L2 error 0.010807823 . Classified as a cat.\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "IEmQmr6_n6dp",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "f0374fa1-1067-4996-f72e-c8afd5a40bed"
      },
      "source": [
        "# only test this on 100 images becuase it takes so long\n",
        "attack = HopSkipJump(classifier=classifier, targeted=False, max_iter=50, max_eval=1000, init_eval=10)\n",
        "correct = 0\n",
        "total = 0\n",
        "for images, labels in test_loader:\n",
        "  target_image = images.numpy()\n",
        "  x_adv = attack.generate(x=target_image, x_adv_init=None, resume=True)\n",
        "  predicted = torch.tensor([np.argmax(classifier.predict(x_adv))])\n",
        "  total += labels.size(0)\n",
        "  correct += (predicted == labels).sum().item()\n",
        "  if total > 100:\n",
        "    break\n",
        "    \n",
        "print('Accuracy of the network against 100 adversarial attack images: %d %%' % (\n",
        "    100 * correct / total))"
      ],
      "execution_count": 10,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Accuracy of the network against 100 adversarial attack images: 10 %\n"
          ],
          "name": "stdout"
        }
      ]
    }
  ]
}