cnn_hnn_adv_attacks_comparsion.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/ericyoc/cnn_hnn_comparison_analysis/blob/main/cnn_hnn_adv_attacks_comparsion.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 1,
      "metadata": {
        "id": "HjqOVmCtH-Is"
      },
      "outputs": [],
      "source": [
        "#!pip install torchattacks"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 2,
      "metadata": {
        "id": "DaENEpfd_BRn"
      },
      "outputs": [],
      "source": [
        "#pip install cirq"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 3,
      "metadata": {
        "id": "bqXmVrAwJG6X"
      },
      "outputs": [],
      "source": [
        "#!pip install tabulate"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 4,
      "metadata": {
        "id": "fpIxGY1oJFS4"
      },
      "outputs": [],
      "source": [
        "from tabulate import tabulate"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 5,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "uQS_bsMUA056",
        "outputId": "70448ae5-d2d5-45c0-bf16-1c44df5e730c"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Drive already mounted at /content/drive; to attempt to forcibly remount, call drive.mount(\"/content/drive\", force_remount=True).\n"
          ]
        }
      ],
      "source": [
        "from google.colab import drive\n",
        "\n",
        "# Mount Google Drive\n",
        "drive.mount('/content/drive')\n",
        "\n",
        "# Access the mounted drive\n",
        "drive_path = '/content/drive/My Drive/'\n",
        "\n",
        "data_dir = drive_path + 'data'"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "import torch\n",
        "import torch.nn as nn\n",
        "import torch.nn.functional as F\n",
        "import torch.optim as optim\n",
        "from torchvision import datasets, transforms\n",
        "from torch.utils.data import DataLoader\n",
        "import numpy as np\n",
        "import math\n",
        "import cirq\n",
        "import os\n",
        "import matplotlib.pyplot as plt\n",
        "import torchattacks\n",
        "from itertools import permutations\n",
        "import copy\n",
        "from tabulate import tabulate\n",
        "\n",
        "# Define the CNN model\n",
        "class CNN(nn.Module):\n",
        "    def __init__(self, device):\n",
        "        super().__init__()\n",
        "        self.conv1 = nn.Conv2d(1, 4, 5)\n",
        "        self.conv2 = nn.Conv2d(4, 10, 5)\n",
        "        self.fc1 = nn.Linear(160, 80)\n",
        "        self.fc2 = nn.Linear(80, 10)\n",
        "        self.device = device\n",
        "        self.to(device)\n",
        "        self._training = True\n",
        "\n",
        "    def forward(self, x):\n",
        "        x = F.relu(self.conv1(x))\n",
        "        x = F.max_pool2d(x, 2)\n",
        "        x = F.relu(self.conv2(x))\n",
        "        x = F.max_pool2d(x, 2)\n",
        "        x = x.view(-1, 160)\n",
        "        x = F.relu(self.fc1(x))\n",
        "        x = self.fc2(x)\n",
        "        return F.log_softmax(x, dim=1)\n",
        "\n",
        "    @property\n",
        "    def shape(self):\n",
        "        return (1, 28, 28)\n",
        "\n",
        "    def clone(self):\n",
        "        return copy.deepcopy(self)\n",
        "\n",
        "    @property\n",
        "    def num_classes(self):\n",
        "        return 10\n",
        "\n",
        "    @property\n",
        "    def training(self):\n",
        "        return self._training\n",
        "\n",
        "    @training.setter\n",
        "    def training(self, mode):\n",
        "        self._training = mode\n",
        "\n",
        "    def detach(self):\n",
        "        for param in self.parameters():\n",
        "            param.detach_()\n",
        "        return self\n",
        "\n",
        "# Create the quantum circuit\n",
        "def create_quantum_circuit(theta, phi, output_dim):\n",
        "    num_qubits = int(math.ceil(math.log2(output_dim)))\n",
        "    qubits = cirq.LineQubit.range(num_qubits)\n",
        "    circuit = cirq.Circuit()\n",
        "    for i, qubit in enumerate(qubits):\n",
        "        if i % 2 == 0:\n",
        "            circuit.append(cirq.ry(float(theta))(qubit))\n",
        "        else:\n",
        "            circuit.append(cirq.ry(float(phi))(qubit))\n",
        "    for i in range(num_qubits - 1):\n",
        "        circuit.append(cirq.CNOT(qubits[i], qubits[i+1]))\n",
        "    return circuit\n",
        "\n",
        "# Simulate the quantum circuit\n",
        "def simulate_circuit(circuit, device):\n",
        "    simulator = cirq.Simulator()\n",
        "    result = simulator.simulate(circuit)\n",
        "    final_state_vector = result.final_state_vector\n",
        "    return final_state_vector\n",
        "\n",
        "# Define the hybrid forward pass\n",
        "def hybrid_forward(input_data, classical_model, theta, phi, device, output_dim=10):\n",
        "    input_data = input_data.to(device)\n",
        "    classical_output = classical_model(input_data)\n",
        "    quantum_circuit = create_quantum_circuit(theta, phi, output_dim)\n",
        "    simulator = cirq.Simulator()\n",
        "    result = simulator.simulate(quantum_circuit)\n",
        "    quantum_output_amplitudes = result.final_state_vector\n",
        "    batch_size = classical_output.size(0)\n",
        "    classical_output = classical_output.view(batch_size, -1)\n",
        "    num_qubits = int(math.ceil(math.log2(output_dim)))\n",
        "    expected_quantum_output_size = 2 ** num_qubits\n",
        "    quantum_output_probabilities = np.square(np.abs(quantum_output_amplitudes))\n",
        "    quantum_output_probabilities = quantum_output_probabilities.reshape(expected_quantum_output_size)[:output_dim]\n",
        "    quantum_output_probabilities = np.tile(quantum_output_probabilities, (batch_size, 1))\n",
        "    quantum_output_probabilities = torch.from_numpy(quantum_output_probabilities).to(device)\n",
        "    alpha = 0.5\n",
        "    beta = 0.5\n",
        "    hybrid_output = alpha * classical_output + beta * quantum_output_probabilities\n",
        "    hybrid_output = F.log_softmax(hybrid_output, dim=1)\n",
        "    return hybrid_output\n",
        "\n",
        "# Define the HNN model\n",
        "class HNN(nn.Module):\n",
        "    def __init__(self, classical_model, device, output_dim=10):\n",
        "        super(HNN, self).__init__()\n",
        "        self.classical_model = classical_model\n",
        "        self.output_dim = output_dim\n",
        "        self.theta = nn.Parameter(torch.randn(1))\n",
        "        self.phi = nn.Parameter(torch.randn(1))\n",
        "        self.device = device\n",
        "        if device is not None:\n",
        "            self.to(device)\n",
        "\n",
        "    def forward(self, input_data):\n",
        "        return hybrid_forward(input_data, self.classical_model, self.theta, self.phi, self.device, self.output_dim)\n",
        "\n",
        "# Load the dataset\n",
        "data_dir = 'data'  # Update with the correct data directory path\n",
        "train_set = datasets.MNIST(data_dir, train=True, download=True, transform=transforms.ToTensor())\n",
        "test_set = datasets.MNIST(data_dir, train=False, download=True, transform=transforms.ToTensor())\n",
        "\n",
        "# Filter the dataset\n",
        "def filtered_dataset(train_set, test_set, filtered=0):\n",
        "    if filtered == 0:  # digits 0 and 1\n",
        "        train_indexes_0 = torch.where(train_set.targets == 0)[0]\n",
        "        train_indexes_1 = torch.where(train_set.targets == 1)[0]\n",
        "        test_indexes_0 = torch.where(test_set.targets == 0)[0]\n",
        "        test_indexes_1 = torch.where(test_set.targets == 1)[0]\n",
        "\n",
        "        num_to_sample_train = min(len(train_indexes_0), len(train_indexes_1))\n",
        "        num_to_sample_test = min(len(test_indexes_0), len(test_indexes_1))\n",
        "\n",
        "        sampled_train_indexes_0 = train_indexes_0[torch.randperm(len(train_indexes_0))[:num_to_sample_train]]\n",
        "        sampled_train_indexes_1 = train_indexes_1[torch.randperm(len(train_indexes_1))[:num_to_sample_train]]\n",
        "        sampled_test_indexes_0 = test_indexes_0[torch.randperm(len(test_indexes_0))[:num_to_sample_test]]\n",
        "        sampled_test_indexes_1 = test_indexes_1[torch.randperm(len(test_indexes_1))[:num_to_sample_test]]\n",
        "\n",
        "        selected_train_indexes = torch.cat([sampled_train_indexes_0, sampled_train_indexes_1])\n",
        "        selected_test_indexes = torch.cat([sampled_test_indexes_0, sampled_test_indexes_1])\n",
        "\n",
        "    elif filtered == 1:  # digits 0 to 9\n",
        "        train_indexes = []\n",
        "        test_indexes = []\n",
        "        for i in range(10):\n",
        "            train_idx = torch.where(train_set.targets == i)[0]\n",
        "            test_idx = torch.where(test_set.targets == i)[0]\n",
        "\n",
        "            train_indexes.append(train_idx)\n",
        "            test_indexes.append(test_idx)\n",
        "\n",
        "        min_num_samples_train = min(len(idx) for idx in train_indexes)\n",
        "        min_num_samples_test = min(len(idx) for idx in test_indexes)\n",
        "\n",
        "        sampled_train_indexes = []\n",
        "        sampled_test_indexes = []\n",
        "        for train_idx, test_idx in zip(train_indexes, test_indexes):\n",
        "            sampled_train_indexes.append(train_idx[torch.randperm(len(train_idx))[:min_num_samples_train]])\n",
        "            sampled_test_indexes.append(test_idx[torch.randperm(len(test_idx))[:min_num_samples_test]])\n",
        "\n",
        "        selected_train_indexes = torch.cat(sampled_train_indexes)\n",
        "        selected_test_indexes = torch.cat(sampled_test_indexes)\n",
        "\n",
        "    else:\n",
        "        raise ValueError(\"Invalid value for 'filtered'. Choose either 0 or 1.\")\n",
        "\n",
        "    filtered_train_set = torch.utils.data.Subset(train_set, selected_train_indexes)\n",
        "    filtered_test_set = torch.utils.data.Subset(test_set, selected_test_indexes)\n",
        "\n",
        "    return filtered_train_set, filtered_test_set\n",
        "\n",
        "filtered_train_set, filtered_test_set = filtered_dataset(train_set, test_set, filtered=1)\n",
        "\n",
        "# Create data loaders\n",
        "filtered_train_loader = DataLoader(filtered_train_set, batch_size=64, shuffle=True)\n",
        "filtered_test_loader = DataLoader(filtered_test_set, batch_size=64)\n",
        "\n",
        "# Set up the device\n",
        "device = 'cuda' if torch.cuda.is_available() else 'cpu'\n",
        "\n",
        "# Create instances of the CNN and HNN models\n",
        "classical_model = CNN(device).to(device)\n",
        "hybrid_model = HNN(classical_model, device=device, output_dim=10).to(device)\n",
        "\n",
        "# Set up the optimizers and loss function\n",
        "cnn_optimizer = optim.Adam(classical_model.parameters())\n",
        "hnn_optimizer = optim.Adam(hybrid_model.parameters())\n",
        "loss_fn = nn.NLLLoss()\n",
        "\n",
        "# Train the CNN\n",
        "def train_cnn(model, train_loader, loss_fn, optimizer, num_epochs, device):\n",
        "    train_accuracies = []\n",
        "    train_losses = []\n",
        "    model.to(device)\n",
        "\n",
        "    for epoch in range(num_epochs):\n",
        "        print(f\"Epoch {epoch+1}/{num_epochs}\")\n",
        "        model.train()\n",
        "        train_loss = 0.0\n",
        "        train_correct = 0\n",
        "        train_total = 0\n",
        "\n",
        "        for i, (imgs, labels) in enumerate(train_loader):\n",
        "            imgs, labels = imgs.to(device), labels.to(device)\n",
        "            preds = model(imgs)\n",
        "            loss = loss_fn(preds, labels)\n",
        "            optimizer.zero_grad()\n",
        "            loss.backward()\n",
        "            optimizer.step()\n",
        "            train_loss += loss.item()\n",
        "            _, predicted = torch.max(preds.data, 1)\n",
        "            train_total += labels.size(0)\n",
        "            train_correct += (predicted == labels).sum().item()\n",
        "\n",
        "        train_loss /= len(train_loader)\n",
        "        train_acc = train_correct / train_total\n",
        "        print(f\"Train Loss: {train_loss:.4f}, Train Accuracy: {train_acc:.4f}\")\n",
        "        train_accuracies.append(train_acc)\n",
        "        train_losses.append(train_loss)\n",
        "\n",
        "    return train_accuracies, train_losses\n",
        "\n",
        "# Train the HNN\n",
        "def train_hnn(model, train_loader, loss_fn, optimizer, num_epochs, device):\n",
        "    train_accuracies = []\n",
        "    train_losses = []\n",
        "    model.to(device)\n",
        "\n",
        "    for epoch in range(num_epochs):\n",
        "        print(f\"Epoch {epoch+1}/{num_epochs}\")\n",
        "        model.train()\n",
        "        train_loss = 0.0\n",
        "        train_correct = 0\n",
        "        train_total = 0\n",
        "\n",
        "        for i, (imgs, labels) in enumerate(train_loader):\n",
        "            imgs, labels = imgs.to(device), labels.to(device)\n",
        "            preds = model(imgs)\n",
        "            loss = loss_fn(preds, labels)\n",
        "            optimizer.zero_grad()\n",
        "            loss.backward()\n",
        "            optimizer.step()\n",
        "            train_loss += loss.item()\n",
        "            _, predicted = torch.max(preds.data, 1)\n",
        "            train_total += labels.size(0)\n",
        "            train_correct += (predicted == labels).sum().item()\n",
        "\n",
        "        train_loss /= len(train_loader)\n",
        "        train_acc = train_correct / train_total\n",
        "        print(f\"Train Loss: {train_loss:.4f}, Train Accuracy: {train_acc:.4f}\")\n",
        "        train_accuracies.append(train_acc)\n",
        "        train_losses.append(train_loss)\n",
        "\n",
        "    return train_accuracies, train_losses\n",
        "\n",
        "# Evaluate the models\n",
        "def evaluate(model, data_loader, loss_fn, device):\n",
        "    test_loss = 0.0\n",
        "    test_accuracy = 0.0\n",
        "    model.eval()\n",
        "\n",
        "    with torch.no_grad():\n",
        "        for i, (x, y) in enumerate(data_loader):\n",
        "            x, y = x.to(device), y.to(device)\n",
        "            logits = model(x)\n",
        "            batch_loss = loss_fn(logits, y)\n",
        "            test_loss += batch_loss.item()\n",
        "            preds = torch.argmax(logits, dim=1)\n",
        "            test_accuracy += torch.sum(preds == y).item() / len(y)\n",
        "\n",
        "    test_loss /= len(data_loader)\n",
        "    test_accuracy /= len(data_loader)\n",
        "    return test_loss, test_accuracy\n",
        "\n",
        "def perform_attack(model, data, labels, attack):\n",
        "    adv_data = attack(data, labels)\n",
        "    return adv_data\n",
        "\n",
        "def print_attack_results(model, test_loader, loss_fn, device, attack_list):\n",
        "    headers = ['Attack', 'Pre-Attack CNN Acc (%)', 'Pre-Attack CNN Loss', 'Post-Attack CNN Acc (%)', 'Post-Attack CNN Loss', 'Pre-Attack HNN Acc (%)', 'Pre-Attack HNN Loss', 'Post-Attack HNN Acc (%)', 'Post-Attack HNN Loss']\n",
        "    table = []\n",
        "\n",
        "    # Get pre-attack accuracy and loss\n",
        "    cnn_test_loss, cnn_test_accuracy = evaluate(classical_model, filtered_test_loader, loss_fn, device)\n",
        "    hnn_test_loss, hnn_test_accuracy = evaluate(hybrid_model, filtered_test_loader, loss_fn, device)\n",
        "\n",
        "    for attack in attack_list:\n",
        "        attack_name = type(attack).__name__\n",
        "        attack_components = attack_name.split('+')\n",
        "        attack_description = ' + '.join(attack_components)\n",
        "\n",
        "        # Evaluate pre-attack performance\n",
        "        cnn_pre_attack_acc = f\"{cnn_test_accuracy * 100:.2f}\"\n",
        "        cnn_pre_attack_loss = f\"{cnn_test_loss:.2f}\"\n",
        "        hnn_pre_attack_acc = f\"{hnn_test_accuracy * 100:.2f}\"\n",
        "        hnn_pre_attack_loss = f\"{hnn_test_loss:.2f}\"\n",
        "\n",
        "        # Perform attack\n",
        "        for data, labels in filtered_test_loader:\n",
        "            data, labels = data.to(device), labels.to(device)\n",
        "            cnn_adv_data = perform_attack(classical_model, data, labels, attack)\n",
        "            hnn_adv_data = perform_attack(hybrid_model, data, labels, attack)\n",
        "\n",
        "            # Evaluate post-attack performance\n",
        "            cnn_post_attack_loss, cnn_post_attack_acc = evaluate(classical_model, [(cnn_adv_data, labels)], loss_fn, device)\n",
        "            hnn_post_attack_loss, hnn_post_attack_acc = evaluate(hybrid_model, [(hnn_adv_data, labels)], loss_fn, device)\n",
        "\n",
        "        row = [attack_description, cnn_pre_attack_acc, cnn_pre_attack_loss, f\"{cnn_post_attack_acc * 100:.2f}\", f\"{cnn_post_attack_loss:.2f}\", hnn_pre_attack_acc, hnn_pre_attack_loss, f\"{hnn_post_attack_acc * 100:.2f}\", f\"{hnn_post_attack_loss:.2f}\"]\n",
        "        table.append(row)\n",
        "\n",
        "    print(tabulate(table, headers=headers, floatfmt=\".2f\"))\n",
        "\n",
        "\n",
        "def main():\n",
        "    # Train the models\n",
        "    num_epochs = 10\n",
        "    cnn_train_accuracies, cnn_train_losses = train_cnn(classical_model, filtered_train_loader, loss_fn, cnn_optimizer, num_epochs, device)\n",
        "    hnn_train_accuracies, hnn_train_losses = train_hnn(hybrid_model, filtered_train_loader, loss_fn, hnn_optimizer, num_epochs, device)\n",
        "\n",
        "    # Evaluate the models\n",
        "    cnn_test_loss, cnn_test_accuracy = evaluate(classical_model, filtered_test_loader, loss_fn, device)\n",
        "    hnn_test_loss, hnn_test_accuracy = evaluate(hybrid_model, filtered_test_loader, loss_fn, device)\n",
        "\n",
        "    print(f\"CNN Test Loss: {cnn_test_loss:.4f}, CNN Test Accuracy: {cnn_test_accuracy:.4f}\")\n",
        "    print(f\"HNN Test Loss: {hnn_test_loss:.4f}, HNN Test Accuracy: {hnn_test_accuracy:.4f}\")\n",
        "\n",
        "    # Print a sample of images with different original and predicted labels\n",
        "    sample_size = 6\n",
        "    misclassified_images = []\n",
        "    with torch.no_grad():\n",
        "        for i, (x, y) in enumerate(filtered_test_loader):\n",
        "            x, y = x.to(device), y.to(device)\n",
        "            cnn_preds = classical_model(x)\n",
        "            hnn_preds = hybrid_model(x)\n",
        "            cnn_preds = torch.argmax(cnn_preds, dim=1)\n",
        "            hnn_preds = torch.argmax(hnn_preds, dim=1)\n",
        "            misclassified_cnn = (cnn_preds != y)\n",
        "            misclassified_hnn = (hnn_preds != y)\n",
        "            misclassified_images.extend([(x[i], y[i], cnn_preds[i], hnn_preds[i]) for i in range(len(x)) if misclassified_cnn[i] or misclassified_hnn[i]])\n",
        "            if len(misclassified_images) >= sample_size:\n",
        "                break\n",
        "\n",
        "    if misclassified_images:\n",
        "        print(\"Sample of misclassified images:\")\n",
        "        for img, label, cnn_pred, hnn_pred in misclassified_images[:sample_size]:\n",
        "            print(f\"Original label: {label}, CNN prediction: {cnn_pred}, HNN prediction: {hnn_pred}\")\n",
        "            if len(img.shape) == 3:\n",
        "                plt.imshow(img.cpu().permute(1, 2, 0), cmap='gray')\n",
        "            else:\n",
        "                plt.imshow(img.cpu().squeeze(), cmap='gray')\n",
        "            plt.show()\n",
        "    else:\n",
        "        print(\"No misclassified images found.\")\n",
        "\n",
        "    # Create the list of individual attacks\n",
        "    attacks = [\n",
        "        \"GN\",\n",
        "        \"FGSM\",\n",
        "        \"BIM\",\n",
        "        \"CW\",\n",
        "        \"RFGSM\",\n",
        "        \"PGD (Linf)\",\n",
        "        \"PGD (L2)\",\n",
        "        \"MIFGSM\",\n",
        "        \"TPGD\",\n",
        "        \"EOTPGD\",\n",
        "        \"APGD\",\n",
        "        \"DIFGSM (Linf)\",\n",
        "        \"Jitter (Linf)\"\n",
        "    ]\n",
        "\n",
        "    headers = ['Attack', 'Pre-Attack CNN Acc (%)', 'Pre-Attack CNN Loss', 'Post-Attack CNN Acc (%)', 'Post-Attack CNN Loss', 'Pre-Attack HNN Acc (%)', 'Pre-Attack HNN Loss', 'Post-Attack HNN Acc (%)', 'Post-Attack HNN Loss']\n",
        "    table = []\n",
        "\n",
        "    # Get pre-attack accuracy and loss\n",
        "    cnn_test_loss, cnn_test_accuracy = evaluate(classical_model, filtered_test_loader, loss_fn, device)\n",
        "    hnn_test_loss, hnn_test_accuracy = evaluate(hybrid_model, filtered_test_loader, loss_fn, device)\n",
        "\n",
        "    compound_attacks = []\n",
        "    for attack1_name, attack2_name in permutations(attacks, 2):\n",
        "        try:\n",
        "\n",
        "            if attack1_name == \"GN\":\n",
        "                cnn_attack1 = torchattacks.GN(classical_model, std=0.1)\n",
        "                hnn_attack1 = torchattacks.GN(hybrid_model, std=0.1)\n",
        "            elif attack1_name == \"FGSM\":\n",
        "                cnn_attack1 = torchattacks.FGSM(classical_model, eps=8/255)\n",
        "                hnn_attack1 = torchattacks.FGSM(hybrid_model, eps=8/255)\n",
        "            elif attack1_name == \"BIM\":\n",
        "                cnn_attack1 = torchattacks.BIM(classical_model, eps=8/255, alpha=2/255, steps=10)\n",
        "                hnn_attack1 = torchattacks.BIM(hybrid_model, eps=8/255, alpha=2/255, steps=10)\n",
        "            elif attack1_name == \"RFGSM\":\n",
        "                cnn_attack1 = torchattacks.BIM(classical_model, eps=8/255, alpha=2/255, steps=10)\n",
        "                hnn_attack1 = torchattacks.BIM(hybrid_model, eps=8/255, alpha=2/255, steps=10)\n",
        "            elif attack1_name == \"PGD (Linf)\":\n",
        "                cnn_attack1 = torchattacks.PGD(classical_model, eps=8/255, alpha=1/255, steps=10, random_start=True)\n",
        "                hnn_attack1 = torchattacks.PGD(hybrid_model, eps=8/255, alpha=1/255, steps=10, random_start=True)\n",
        "            elif attack1_name == \"PGD (L2)\":\n",
        "                cnn_attack1 = torchattacks.PGDL2(classical_model, eps=1.0, alpha=0.2, steps=10, random_start=True)\n",
        "                hnn_attack1 = torchattacks.PGDL2(hybrid_model, eps=1.0, alpha=0.2, steps=10, random_start=True)\n",
        "            elif attack1_name == \"MIFGSM\":\n",
        "                cnn_attack1 = torchattacks.MIFGSM(classical_model, eps=8/255, steps=10, decay=1.0)\n",
        "                hnn_attack1 = torchattacks.MIFGSM(hybrid_model, eps=8/255, steps=10, decay=1.0)\n",
        "            elif attack1_name == \"TPGD\":\n",
        "                cnn_attack1 = torchattacks.TPGD(classical_model, eps=8/255, alpha=2/255, steps=10)\n",
        "                hnn_attack1 = torchattacks.TPGD(hybrid_model, eps=8/255, alpha=2/255, steps=10)\n",
        "            elif attack1_name == \"EOTPGD\":\n",
        "                cnn_attack1 = torchattacks.EOTPGD(classical_model, eps=8/255, alpha=2/255, steps=10, eot_iter=2)\n",
        "                hnn_attack1 = torchattacks.EOTPGD(hybrid_model, eps=8/255, alpha=2/255, steps=10, eot_iter=2)\n",
        "            elif attack1_name == \"APGD\":\n",
        "                cnn_attack1 = torchattacks.APGD(classical_model, norm='Linf', eps=8/255, steps=10, n_restarts=1, seed=0, loss='ce', eot_iter=1, rho=.75, verbose=False)\n",
        "                hnn_attack1 = torchattacks.APGD(hybrid_model, norm='Linf', eps=8/255, steps=10, n_restarts=1, seed=0, loss='ce', eot_iter=1, rho=.75, verbose=False)\n",
        "            elif attack1_name == \"DIFGSM (Linf)\":\n",
        "                cnn_attack1 = torchattacks.DIFGSM(classical_model, eps=8/255, alpha=2/255, steps=10, decay=0.0, resize_rate=0.9, diversity_prob=0.5, random_start=False)\n",
        "                hnn_attack1 = torchattacks.DIFGSM(hybrid_model, eps=8/255, alpha=2/255, steps=10, decay=0.0, resize_rate=0.9, diversity_prob=0.5, random_start=False)\n",
        "            elif attack1_name == \"Jitter (Linf)\":\n",
        "                cnn_attack1 = torchattacks.Jitter(classical_model, eps=8/255, alpha=2/255, steps=10, scale=10, std=0.1, random_start=True)\n",
        "                hnn_attack1 = torchattacks.Jitter(hybrid_model, eps=8/255, alpha=2/255, steps=10, scale=10, std=0.1, random_start=True)\n",
        "            elif attack1_name == \"CW\":\n",
        "                cnn_attack1 = torchattacks.CW(classical_model, c=1, kappa=0, steps=200, lr=0.01)\n",
        "                hnn_attack1 = torchattacks.CW(hybrid_model, c=1, kappa=0, steps=200, lr=0.01)\n",
        "            else:\n",
        "                raise ValueError(f\"Unsupported attack type: {attack1_name}\")\n",
        "\n",
        "            if attack2_name == \"GN\":\n",
        "                cnn_attack2 = torchattacks.GN(classical_model, std=0.1)\n",
        "                hnn_attack2 = torchattacks.GN(hybrid_model, std=0.1)\n",
        "            elif attack2_name == \"FGSM\":\n",
        "                cnn_attack2 = torchattacks.FGSM(classical_model, eps=8/255)\n",
        "                hnn_attack2 = torchattacks.FGSM(hybrid_model, eps=8/255)\n",
        "            elif attack2_name == \"BIM\":\n",
        "                cnn_attack2 = torchattacks.BIM(classical_model, eps=8/255, alpha=2/255, steps=10)\n",
        "                hnn_attack2 = torchattacks.BIM(hybrid_model, eps=8/255, alpha=2/255, steps=10)\n",
        "            elif attack2_name == \"RFGSM\":\n",
        "                cnn_attack2 = torchattacks.BIM(classical_model, eps=8/255, alpha=2/255, steps=10)\n",
        "                hnn_attack2 = torchattacks.BIM(hybrid_model, eps=8/255, alpha=2/255, steps=10)\n",
        "            elif attack2_name == \"PGD (Linf)\":\n",
        "                cnn_attack2 = torchattacks.PGD(classical_model, eps=8/255, alpha=1/255, steps=10, random_start=True)\n",
        "                hnn_attack2 = torchattacks.PGD(hybrid_model, eps=8/255, alpha=1/255, steps=10, random_start=True)\n",
        "            elif attack2_name == \"PGD (L2)\":\n",
        "                cnn_attack2 = torchattacks.PGDL2(classical_model, eps=1.0, alpha=0.2, steps=10, random_start=True)\n",
        "                hnn_attack2 = torchattacks.PGDL2(hybrid_model, eps=1.0, alpha=0.2, steps=10, random_start=True)\n",
        "            elif attack2_name == \"MIFGSM\":\n",
        "                cnn_attack2 = torchattacks.MIFGSM(classical_model, eps=8/255, steps=10, decay=1.0)\n",
        "                hnn_attack2 = torchattacks.MIFGSM(hybrid_model, eps=8/255, steps=10, decay=1.0)\n",
        "            elif attack2_name == \"TPGD\":\n",
        "                cnn_attack2 = torchattacks.TPGD(classical_model, eps=8/255, alpha=2/255, steps=10)\n",
        "                hnn_attack2 = torchattacks.TPGD(hybrid_model, eps=8/255, alpha=2/255, steps=10)\n",
        "            elif attack2_name == \"EOTPGD\":\n",
        "                cnn_attack2 = torchattacks.EOTPGD(classical_model, eps=8/255, alpha=2/255, steps=10, eot_iter=2)\n",
        "                hnn_attack2 = torchattacks.EOTPGD(hybrid_model, eps=8/255, alpha=2/255, steps=10, eot_iter=2)\n",
        "            elif attack2_name == \"APGD\":\n",
        "                cnn_attack2 = torchattacks.APGD(classical_model, norm='Linf', eps=8/255, steps=10, n_restarts=1, seed=0, loss='ce', eot_iter=1, rho=.75, verbose=False)\n",
        "                hnn_attack2 = torchattacks.APGD(hybrid_model, norm='Linf', eps=8/255, steps=10, n_restarts=1, seed=0, loss='ce', eot_iter=1, rho=.75, verbose=False)\n",
        "            elif attack2_name == \"DIFGSM (Linf)\":\n",
        "                cnn_attack2 = torchattacks.DIFGSM(classical_model, eps=8/255, alpha=2/255, steps=10, decay=0.0, resize_rate=0.9, diversity_prob=0.5, random_start=False)\n",
        "                hnn_attack2 = torchattacks.DIFGSM(hybrid_model, eps=8/255, alpha=2/255, steps=10, decay=0.0, resize_rate=0.9, diversity_prob=0.5, random_start=False)\n",
        "            elif attack2_name == \"Jitter (Linf)\":\n",
        "                cnn_attack2 = torchattacks.Jitter(classical_model, eps=8/255, alpha=2/255, steps=10, scale=10, std=0.1, random_start=True)\n",
        "                hnn_attack2 = torchattacks.Jitter(hybrid_model, eps=8/255, alpha=2/255, steps=10, scale=10, std=0.1, random_start=True)\n",
        "            elif attack2_name == \"CW\":\n",
        "                cnn_attack2 = torchattacks.CW(classical_model, c=1, kappa=0, steps=200, lr=0.01)\n",
        "                hnn_attack2 = torchattacks.CW(hybrid_model, c=1, kappa=0, steps=200, lr=0.01)\n",
        "            else:\n",
        "                raise ValueError(f\"Unsupported attack type: {attack2_name}\")\n",
        "\n",
        "            cnn_compound_attack = torchattacks.MultiAttack([cnn_attack1, cnn_attack2])\n",
        "            hnn_compound_attack = torchattacks.MultiAttack([hnn_attack1, hnn_attack2])\n",
        "            attack_description = f\"{attack1_name} + {attack2_name}\"\n",
        "\n",
        "            # Evaluate pre-attack performance\n",
        "            cnn_pre_attack_acc = f\"{cnn_test_accuracy * 100:.2f}\"\n",
        "            cnn_pre_attack_loss = f\"{cnn_test_loss:.2f}\"\n",
        "            hnn_pre_attack_acc = f\"{hnn_test_accuracy * 100:.2f}\"\n",
        "            hnn_pre_attack_loss = f\"{hnn_test_loss:.2f}\"\n",
        "\n",
        "            # Perform attack\n",
        "            for data, labels in filtered_test_loader:\n",
        "                data, labels = data.to(device), labels.to(device)\n",
        "                cnn_adv_data = perform_attack(classical_model, data, labels, cnn_compound_attack)\n",
        "                hnn_adv_data = perform_attack(hybrid_model, data, labels, hnn_compound_attack)\n",
        "\n",
        "                # Evaluate post-attack performance\n",
        "                cnn_post_attack_loss, cnn_post_attack_acc = evaluate(classical_model, [(cnn_adv_data, labels)], loss_fn, device)\n",
        "                hnn_post_attack_loss, hnn_post_attack_acc = evaluate(hybrid_model, [(hnn_adv_data, labels)], loss_fn, device)\n",
        "\n",
        "            row = [attack_description, cnn_pre_attack_acc, cnn_pre_attack_loss, f\"{cnn_post_attack_acc * 100:.2f}\", f\"{cnn_post_attack_loss:.2f}\", hnn_pre_attack_acc, hnn_pre_attack_loss, f\"{hnn_post_attack_acc * 100:.2f}\", f\"{hnn_post_attack_loss:.2f}\"]\n",
        "            table.append(row)\n",
        "\n",
        "            print(tabulate([row], headers=headers, floatfmt=\".2f\"))\n",
        "\n",
        "        except (AttributeError, ValueError) as e:\n",
        "            print(f\"Error: {e}\")\n",
        "\n",
        "    print(\"\\nFinal Results:\")\n",
        "    print(tabulate(table, headers=headers, floatfmt=\".2f\"))\n",
        "\n",
        "\n",
        "    #Find the top three compound attacks with the greatest accuracy difference for CNN and HNN\n",
        "    cnn_accuracy_diffs = [(row[0], float(row[1]) - float(row[3])) for row in table]\n",
        "    hnn_accuracy_diffs = [(row[0], float(row[5]) - float(row[7])) for row in table]\n",
        "\n",
        "    cnn_top_attacks = sorted(cnn_accuracy_diffs, key=lambda x: x[1], reverse=True)[:3]\n",
        "    hnn_top_attacks = sorted(hnn_accuracy_diffs, key=lambda x: x[1], reverse=True)[:3]\n",
        "\n",
        "    cnn_top_table = []\n",
        "    hnn_top_table = []\n",
        "\n",
        "    for attack, diff in cnn_top_attacks:\n",
        "        row = [attack, f\"{diff:.2f}\"]\n",
        "        cnn_top_table.append(row)\n",
        "\n",
        "    for attack, diff in hnn_top_attacks:\n",
        "        row = [attack, f\"{diff:.2f}\"]\n",
        "        hnn_top_table.append(row)\n",
        "\n",
        "    print(\"\\nTop 3 Compound Attacks with Greatest Accuracy Difference for CNN:\")\n",
        "    print(tabulate(cnn_top_table, headers=[\"Attack\", \"Accuracy Difference (%)\"], floatfmt=\".2f\"))\n",
        "\n",
        "    print(\"\\nTop 3 Compound Attacks with Greatest Accuracy Difference for HNN:\")\n",
        "    print(tabulate(hnn_top_table, headers=[\"Attack\", \"Accuracy Difference (%)\"], floatfmt=\".2f\"))\n",
        "\n",
        "    # Plot the accuracy and loss curves\n",
        "    epochs = range(1, num_epochs + 1)\n",
        "\n",
        "    plt.figure(figsize=(12, 6))\n",
        "    plt.subplot(1, 2, 1)\n",
        "    plt.plot(epochs, cnn_train_accuracies, 'b', label='CNN Training Accuracy')\n",
        "    plt.plot(epochs, hnn_train_accuracies, 'r', label='HNN Training Accuracy')\n",
        "    plt.title('Training Accuracy')\n",
        "    plt.xlabel('Epochs')\n",
        "    plt.ylabel('Accuracy')\n",
        "    plt.legend()\n",
        "\n",
        "    plt.subplot(1, 2, 2)\n",
        "    plt.plot(epochs, cnn_train_losses, 'b', label='CNN Training Loss')\n",
        "    plt.plot(epochs, hnn_train_losses, 'r', label='HNN Training Loss')\n",
        "    plt.title('Training Loss')\n",
        "    plt.xlabel('Epochs')\n",
        "    plt.ylabel('Loss')\n",
        "    plt.legend()\n",
        "\n",
        "    plt.tight_layout()\n",
        "    plt.show()\n",
        "\n",
        "if __name__ == \"__main__\":\n",
        "    main()\n",
        ""
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "id": "Zi8K3Xk7j8nV",
        "outputId": "3f6c9b82-9ea5-4b11-84a8-4dc14e8a71ee"
      },
      "execution_count": 13,
      "outputs": [
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Epoch 1/10\n",
            "Train Loss: 0.3639, Train Accuracy: 0.8935\n",
            "Epoch 2/10\n",
            "Train Loss: 0.0998, Train Accuracy: 0.9699\n",
            "Epoch 3/10\n",
            "Train Loss: 0.0713, Train Accuracy: 0.9784\n",
            "Epoch 4/10\n",
            "Train Loss: 0.0615, Train Accuracy: 0.9821\n",
            "Epoch 5/10\n",
            "Train Loss: 0.0508, Train Accuracy: 0.9844\n",
            "Epoch 6/10\n",
            "Train Loss: 0.0422, Train Accuracy: 0.9865\n",
            "Epoch 7/10\n",
            "Train Loss: 0.0378, Train Accuracy: 0.9878\n",
            "Epoch 8/10\n",
            "Train Loss: 0.0332, Train Accuracy: 0.9895\n",
            "Epoch 9/10\n",
            "Train Loss: 0.0288, Train Accuracy: 0.9905\n",
            "Epoch 10/10\n",
            "Train Loss: 0.0253, Train Accuracy: 0.9919\n",
            "Epoch 1/10\n",
            "Train Loss: 0.0281, Train Accuracy: 0.9910\n",
            "Epoch 2/10\n",
            "Train Loss: 0.0234, Train Accuracy: 0.9924\n",
            "Epoch 3/10\n",
            "Train Loss: 0.0187, Train Accuracy: 0.9941\n",
            "Epoch 4/10\n",
            "Train Loss: 0.0175, Train Accuracy: 0.9948\n",
            "Epoch 5/10\n",
            "Train Loss: 0.0162, Train Accuracy: 0.9948\n",
            "Epoch 6/10\n",
            "Train Loss: 0.0148, Train Accuracy: 0.9954\n",
            "Epoch 7/10\n",
            "Train Loss: 0.0137, Train Accuracy: 0.9952\n",
            "Epoch 8/10\n",
            "Train Loss: 0.0114, Train Accuracy: 0.9962\n",
            "Epoch 9/10\n",
            "Train Loss: 0.0107, Train Accuracy: 0.9964\n",
            "Epoch 10/10\n",
            "Train Loss: 0.0093, Train Accuracy: 0.9970\n",
            "CNN Test Loss: 0.0656, CNN Test Accuracy: 0.9903\n",
            "HNN Test Loss: 0.0379, HNN Test Accuracy: 0.9903\n",
            "Sample of misclassified images:\n",
            "Original label: 0, CNN prediction: 8, HNN prediction: 8\n"
          ]
        },
        {
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Original label: 0, CNN prediction: 9, HNN prediction: 9\n"
          ]
        },
        {
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Original label: 0, CNN prediction: 6, HNN prediction: 6\n"
          ]
        },
        {
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Original label: 0, CNN prediction: 1, HNN prediction: 1\n"
          ]
        },
        {
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Original label: 0, CNN prediction: 2, HNN prediction: 2\n"
          ]
        },
        {
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "metadata": {
            "tags": null
          },
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Original label: 0, CNN prediction: 7, HNN prediction: 7\n"
          ]
        },
        {
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Attack       Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "GN + FGSM                     99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.28\n",
            "Attack      Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "GN + BIM                     99.03                   0.07                      87.50                    0.68                     99.03                   0.04                      87.50                    0.51\n",
            "Attack      Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "GN + CW                      99.03                   0.07                     100.00                    0.00                     99.03                   0.04                     100.00                    0.02\n",
            "Attack        Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "GN + RFGSM                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.34\n",
            "Attack             Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "GN + PGD (Linf)                     99.03                   0.07                      87.50                    0.87                     99.03                   0.04                      87.50                    0.46\n",
            "Attack           Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "GN + PGD (L2)                     99.03                   0.07                      45.83                    4.58                     99.03                   0.04                      37.50                    3.19\n",
            "Attack         Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "GN + MIFGSM                     99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "Attack       Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "GN + TPGD                     99.03                   0.07                      91.67                    0.51                     99.03                   0.04                      91.67                    0.28\n",
            "Attack         Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "GN + EOTPGD                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "Attack       Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "GN + APGD                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "Attack                Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "GN + DIFGSM (Linf)                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      91.67                    0.37\n",
            "Attack                Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "GN + Jitter (Linf)                     99.03                   0.07                      87.50                    0.55                     99.03                   0.04                      87.50                    0.50\n",
            "Attack       Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "FGSM + GN                     99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "Attack        Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "FGSM + BIM                     99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "Attack       Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "FGSM + CW                     99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "Attack          Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "FGSM + RFGSM                     99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "Attack               Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "FGSM + PGD (Linf)                     99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "Attack             Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "FGSM + PGD (L2)                     99.03                   0.07                      41.67                    3.62                     99.03                   0.04                      37.50                    2.07\n",
            "Attack           Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "FGSM + MIFGSM                     99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "Attack         Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "FGSM + TPGD                     99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "Attack           Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "FGSM + EOTPGD                     99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "Attack         Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "FGSM + APGD                     99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "Attack                  Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "FGSM + DIFGSM (Linf)                     99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "Attack                  Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "FGSM + Jitter (Linf)                     99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "Attack      Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "BIM + GN                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack        Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "BIM + FGSM                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack      Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "BIM + CW                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack         Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "BIM + RFGSM                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack              Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "----------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "BIM + PGD (Linf)                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack            Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "BIM + PGD (L2)                     99.03                   0.07                      45.83                    3.76                     99.03                   0.04                      37.50                    2.20\n",
            "Attack          Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "BIM + MIFGSM                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack        Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "BIM + TPGD                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack          Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "BIM + EOTPGD                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack        Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "BIM + APGD                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack                 Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "BIM + DIFGSM (Linf)                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack                 Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "BIM + Jitter (Linf)                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack      Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "CW + GN                      99.03                   0.07                      95.83                    0.05                     99.03                   0.04                     100.00                    0.02\n",
            "Attack       Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "CW + FGSM                     99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "Attack      Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "CW + BIM                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack        Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "CW + RFGSM                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack             Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "CW + PGD (Linf)                     99.03                   0.07                      87.50                    0.91                     99.03                   0.04                      87.50                    0.47\n",
            "Attack           Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "CW + PGD (L2)                     99.03                   0.07                      45.83                    5.68                     99.03                   0.04                      37.50                    3.20\n",
            "Attack         Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "CW + MIFGSM                     99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "Attack       Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "CW + TPGD                     99.03                   0.07                      91.67                    0.51                     99.03                   0.04                      91.67                    0.28\n",
            "Attack         Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "CW + EOTPGD                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.52\n",
            "Attack       Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "CW + APGD                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "Attack                Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "CW + DIFGSM (Linf)                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      91.67                    0.37\n",
            "Attack                Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "CW + Jitter (Linf)                     99.03                   0.07                      87.50                    0.94                     99.03                   0.04                      87.50                    0.50\n",
            "Attack        Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "RFGSM + GN                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack          Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "RFGSM + FGSM                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack         Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "RFGSM + BIM                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack        Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "RFGSM + CW                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack                Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "RFGSM + PGD (Linf)                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack              Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "----------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "RFGSM + PGD (L2)                     99.03                   0.07                      45.83                    3.76                     99.03                   0.04                      37.50                    2.20\n",
            "Attack            Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "RFGSM + MIFGSM                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack          Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "RFGSM + TPGD                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack            Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "RFGSM + EOTPGD                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack          Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "RFGSM + APGD                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack                   Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "RFGSM + DIFGSM (Linf)                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack                   Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "RFGSM + Jitter (Linf)                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack             Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (Linf) + GN                     99.03                   0.07                      87.50                    0.91                     99.03                   0.04                      87.50                    0.47\n",
            "Attack               Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (Linf) + FGSM                     99.03                   0.07                      87.50                    0.88                     99.03                   0.04                      87.50                    0.45\n",
            "Attack              Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "----------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (Linf) + BIM                     99.03                   0.07                      87.50                    0.91                     99.03                   0.04                      87.50                    0.48\n",
            "Attack             Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (Linf) + CW                     99.03                   0.07                      87.50                    0.89                     99.03                   0.04                      87.50                    0.45\n",
            "Attack                Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (Linf) + RFGSM                     99.03                   0.07                      87.50                    0.92                     99.03                   0.04                      87.50                    0.47\n",
            "Attack                   Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (Linf) + PGD (L2)                     99.03                   0.07                      50.00                    3.56                     99.03                   0.04                      37.50                    2.21\n",
            "Attack                 Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (Linf) + MIFGSM                     99.03                   0.07                      87.50                    0.89                     99.03                   0.04                      87.50                    0.48\n",
            "Attack               Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (Linf) + TPGD                     99.03                   0.07                      87.50                    0.86                     99.03                   0.04                      87.50                    0.46\n",
            "Attack                 Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (Linf) + EOTPGD                     99.03                   0.07                      87.50                    0.89                     99.03                   0.04                      87.50                    0.46\n",
            "Attack               Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (Linf) + APGD                     99.03                   0.07                      87.50                    0.90                     99.03                   0.04                      87.50                    0.48\n",
            "Attack                        Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (Linf) + DIFGSM (Linf)                     99.03                   0.07                      87.50                    0.89                     99.03                   0.04                      87.50                    0.47\n",
            "Attack                        Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (Linf) + Jitter (Linf)                     99.03                   0.07                      87.50                    0.86                     99.03                   0.04                      87.50                    0.48\n",
            "Attack           Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (L2) + GN                     99.03                   0.07                      45.83                    5.74                     99.03                   0.04                      37.50                    3.31\n",
            "Attack             Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (L2) + FGSM                     99.03                   0.07                      45.83                    5.59                     99.03                   0.04                      37.50                    3.06\n",
            "Attack            Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (L2) + BIM                     99.03                   0.07                      45.83                    5.62                     99.03                   0.04                      37.50                    3.33\n",
            "Attack           Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (L2) + CW                     99.03                   0.07                      45.83                    5.69                     99.03                   0.04                      37.50                    3.20\n",
            "Attack              Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "----------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (L2) + RFGSM                     99.03                   0.07                      45.83                    5.69                     99.03                   0.04                      37.50                    3.07\n",
            "Attack                   Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (L2) + PGD (Linf)                     99.03                   0.07                      45.83                    5.62                     99.03                   0.04                      37.50                    3.19\n",
            "Attack               Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (L2) + MIFGSM                     99.03                   0.07                      50.00                    5.71                     99.03                   0.04                      37.50                    3.30\n",
            "Attack             Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (L2) + TPGD                     99.03                   0.07                      50.00                    5.69                     99.03                   0.04                      37.50                    3.23\n",
            "Attack               Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (L2) + EOTPGD                     99.03                   0.07                      45.83                    5.68                     99.03                   0.04                      37.50                    3.28\n",
            "Attack             Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (L2) + APGD                     99.03                   0.07                      45.83                    5.60                     99.03                   0.04                      37.50                    3.23\n",
            "Attack                      Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (L2) + DIFGSM (Linf)                     99.03                   0.07                      45.83                    5.48                     99.03                   0.04                      37.50                    3.34\n",
            "Attack                      Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "PGD (L2) + Jitter (Linf)                     99.03                   0.07                      45.83                    5.57                     99.03                   0.04                      37.50                    3.05\n",
            "Attack         Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "MIFGSM + GN                     99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "Attack           Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "MIFGSM + FGSM                     99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "Attack          Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "MIFGSM + BIM                     99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "Attack         Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "MIFGSM + CW                     99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "Attack            Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "MIFGSM + RFGSM                     99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "Attack                 Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "MIFGSM + PGD (Linf)                     99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "Attack               Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "MIFGSM + PGD (L2)                     99.03                   0.07                      45.83                    3.68                     99.03                   0.04                      37.50                    2.24\n",
            "Attack           Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "MIFGSM + TPGD                     99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "Attack             Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "MIFGSM + EOTPGD                     99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "Attack           Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "MIFGSM + APGD                     99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "Attack                    Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "----------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "MIFGSM + DIFGSM (Linf)                     99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "Attack                    Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "----------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "MIFGSM + Jitter (Linf)                     99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "Attack       Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "TPGD + GN                     99.03                   0.07                      91.67                    0.52                     99.03                   0.04                      87.50                    0.51\n",
            "Attack         Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "TPGD + FGSM                     99.03                   0.07                      87.50                    0.90                     99.03                   0.04                      87.50                    0.46\n",
            "Attack        Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "TPGD + BIM                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.52\n",
            "Attack       Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "TPGD + CW                     99.03                   0.07                      91.67                    0.51                     99.03                   0.04                      91.67                    0.28\n",
            "Attack          Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "TPGD + RFGSM                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.53\n",
            "Attack               Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "TPGD + PGD (Linf)                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.47\n",
            "Attack             Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "TPGD + PGD (L2)                     99.03                   0.07                      45.83                    4.31                     99.03                   0.04                      37.50                    2.51\n",
            "Attack           Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "TPGD + MIFGSM                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.52\n",
            "Attack           Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "TPGD + EOTPGD                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.51\n",
            "Attack         Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "TPGD + APGD                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.51\n",
            "Attack                  Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "TPGD + DIFGSM (Linf)                     99.03                   0.07                      87.50                    0.96                     99.03                   0.04                      87.50                    0.51\n",
            "Attack                  Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "TPGD + Jitter (Linf)                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.50\n",
            "Attack         Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "EOTPGD + GN                     99.03                   0.07                      87.50                    0.97                     99.03                   0.04                      87.50                    0.52\n",
            "Attack           Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "EOTPGD + FGSM                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.52\n",
            "Attack          Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "EOTPGD + BIM                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.52\n",
            "Attack         Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "EOTPGD + CW                     99.03                   0.07                      87.50                    0.97                     99.03                   0.04                      87.50                    0.50\n",
            "Attack            Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "EOTPGD + RFGSM                     99.03                   0.07                      87.50                    0.97                     99.03                   0.04                      87.50                    0.53\n",
            "Attack                 Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "EOTPGD + PGD (Linf)                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.50\n",
            "Attack               Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "EOTPGD + PGD (L2)                     99.03                   0.07                      45.83                    3.43                     99.03                   0.04                      37.50                    2.13\n",
            "Attack             Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "EOTPGD + MIFGSM                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack           Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "EOTPGD + TPGD                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.51\n",
            "Attack           Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "EOTPGD + APGD                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.53\n",
            "Attack                    Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "----------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "EOTPGD + DIFGSM (Linf)                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.53\n",
            "Attack                    Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "----------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "EOTPGD + Jitter (Linf)                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.51\n",
            "Attack       Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "APGD + GN                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "Attack         Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "APGD + FGSM                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "Attack        Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "APGD + BIM                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "Attack       Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "APGD + CW                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "Attack          Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "APGD + RFGSM                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "Attack               Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "APGD + PGD (Linf)                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "Attack             Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "APGD + PGD (L2)                     99.03                   0.07                      45.83                    3.65                     99.03                   0.04                      37.50                    2.17\n",
            "Attack           Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "APGD + MIFGSM                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "Attack         Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "APGD + TPGD                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "Attack           Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "APGD + EOTPGD                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "Attack                  Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "APGD + DIFGSM (Linf)                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "Attack                  Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "APGD + Jitter (Linf)                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "Attack                Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "DIFGSM (Linf) + GN                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      91.67                    0.37\n",
            "Attack                  Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "DIFGSM (Linf) + FGSM                     99.03                   0.07                      87.50                    0.75                     99.03                   0.04                      87.50                    0.38\n",
            "Attack                 Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "DIFGSM (Linf) + BIM                     99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.50\n",
            "Attack                Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "DIFGSM (Linf) + CW                     99.03                   0.07                      91.67                    0.73                     99.03                   0.04                      87.50                    0.49\n",
            "Attack                   Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "DIFGSM (Linf) + RFGSM                     99.03                   0.07                      87.50                    0.94                     99.03                   0.04                      87.50                    0.39\n",
            "Attack                        Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "DIFGSM (Linf) + PGD (Linf)                     99.03                   0.07                      87.50                    0.84                     99.03                   0.04                      87.50                    0.40\n",
            "Attack                      Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "DIFGSM (Linf) + PGD (L2)                     99.03                   0.07                      41.67                    3.62                     99.03                   0.04                      37.50                    2.04\n",
            "Attack                    Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "----------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "DIFGSM (Linf) + MIFGSM                     99.03                   0.07                      87.50                    0.94                     99.03                   0.04                      87.50                    0.34\n",
            "Attack                  Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "DIFGSM (Linf) + TPGD                     99.03                   0.07                      87.50                    0.77                     99.03                   0.04                      87.50                    0.43\n",
            "Attack                    Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "----------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "DIFGSM (Linf) + EOTPGD                     99.03                   0.07                      87.50                    0.81                     99.03                   0.04                      87.50                    0.44\n",
            "Attack                  Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "DIFGSM (Linf) + APGD                     99.03                   0.07                      87.50                    0.72                     99.03                   0.04                      87.50                    0.50\n",
            "Attack                           Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "DIFGSM (Linf) + Jitter (Linf)                     99.03                   0.07                      87.50                    0.72                     99.03                   0.04                      87.50                    0.45\n",
            "Attack                Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "Jitter (Linf) + GN                     99.03                   0.07                      87.50                    0.97                     99.03                   0.04                      87.50                    0.50\n",
            "Attack                  Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "Jitter (Linf) + FGSM                     99.03                   0.07                      87.50                    0.96                     99.03                   0.04                      87.50                    0.50\n",
            "Attack                 Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "Jitter (Linf) + BIM                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.51\n",
            "Attack                Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "Jitter (Linf) + CW                     99.03                   0.07                      87.50                    0.97                     99.03                   0.04                      87.50                    0.49\n",
            "Attack                   Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "---------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "Jitter (Linf) + RFGSM                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.50\n",
            "Attack                        Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "Jitter (Linf) + PGD (Linf)                     99.03                   0.07                      87.50                    0.94                     99.03                   0.04                      87.50                    0.49\n",
            "Attack                      Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "------------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "Jitter (Linf) + PGD (L2)                     99.03                   0.07                      45.83                    3.60                     99.03                   0.04                      37.50                    2.18\n",
            "Attack                    Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "----------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "Jitter (Linf) + MIFGSM                     99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.50\n",
            "Attack                  Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "Jitter (Linf) + TPGD                     99.03                   0.07                      87.50                    0.93                     99.03                   0.04                      87.50                    0.51\n",
            "Attack                    Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "----------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "Jitter (Linf) + EOTPGD                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Attack                  Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "--------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "Jitter (Linf) + APGD                     99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.51\n",
            "Attack                           Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "Jitter (Linf) + DIFGSM (Linf)                     99.03                   0.07                      87.50                    0.94                     99.03                   0.04                      87.50                    0.50\n",
            "\n",
            "Final Results:\n",
            "Attack                           Pre-Attack CNN Acc (%)    Pre-Attack CNN Loss    Post-Attack CNN Acc (%)    Post-Attack CNN Loss    Pre-Attack HNN Acc (%)    Pre-Attack HNN Loss    Post-Attack HNN Acc (%)    Post-Attack HNN Loss\n",
            "-----------------------------  ------------------------  ---------------------  -------------------------  ----------------------  ------------------------  ---------------------  -------------------------  ----------------------\n",
            "GN + FGSM                                         99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.28\n",
            "GN + BIM                                          99.03                   0.07                      87.50                    0.68                     99.03                   0.04                      87.50                    0.51\n",
            "GN + CW                                           99.03                   0.07                     100.00                    0.00                     99.03                   0.04                     100.00                    0.02\n",
            "GN + RFGSM                                        99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.34\n",
            "GN + PGD (Linf)                                   99.03                   0.07                      87.50                    0.87                     99.03                   0.04                      87.50                    0.46\n",
            "GN + PGD (L2)                                     99.03                   0.07                      45.83                    4.58                     99.03                   0.04                      37.50                    3.19\n",
            "GN + MIFGSM                                       99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "GN + TPGD                                         99.03                   0.07                      91.67                    0.51                     99.03                   0.04                      91.67                    0.28\n",
            "GN + EOTPGD                                       99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "GN + APGD                                         99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "GN + DIFGSM (Linf)                                99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      91.67                    0.37\n",
            "GN + Jitter (Linf)                                99.03                   0.07                      87.50                    0.55                     99.03                   0.04                      87.50                    0.50\n",
            "FGSM + GN                                         99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "FGSM + BIM                                        99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "FGSM + CW                                         99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "FGSM + RFGSM                                      99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "FGSM + PGD (Linf)                                 99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "FGSM + PGD (L2)                                   99.03                   0.07                      41.67                    3.62                     99.03                   0.04                      37.50                    2.07\n",
            "FGSM + MIFGSM                                     99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "FGSM + TPGD                                       99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "FGSM + EOTPGD                                     99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "FGSM + APGD                                       99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "FGSM + DIFGSM (Linf)                              99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "FGSM + Jitter (Linf)                              99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "BIM + GN                                          99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "BIM + FGSM                                        99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "BIM + CW                                          99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "BIM + RFGSM                                       99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "BIM + PGD (Linf)                                  99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "BIM + PGD (L2)                                    99.03                   0.07                      45.83                    3.76                     99.03                   0.04                      37.50                    2.20\n",
            "BIM + MIFGSM                                      99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "BIM + TPGD                                        99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "BIM + EOTPGD                                      99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "BIM + APGD                                        99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "BIM + DIFGSM (Linf)                               99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "BIM + Jitter (Linf)                               99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "CW + GN                                           99.03                   0.07                      95.83                    0.05                     99.03                   0.04                     100.00                    0.02\n",
            "CW + FGSM                                         99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.43\n",
            "CW + BIM                                          99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "CW + RFGSM                                        99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "CW + PGD (Linf)                                   99.03                   0.07                      87.50                    0.91                     99.03                   0.04                      87.50                    0.47\n",
            "CW + PGD (L2)                                     99.03                   0.07                      45.83                    5.68                     99.03                   0.04                      37.50                    3.20\n",
            "CW + MIFGSM                                       99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "CW + TPGD                                         99.03                   0.07                      91.67                    0.51                     99.03                   0.04                      91.67                    0.28\n",
            "CW + EOTPGD                                       99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.52\n",
            "CW + APGD                                         99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "CW + DIFGSM (Linf)                                99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      91.67                    0.37\n",
            "CW + Jitter (Linf)                                99.03                   0.07                      87.50                    0.94                     99.03                   0.04                      87.50                    0.50\n",
            "RFGSM + GN                                        99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "RFGSM + FGSM                                      99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "RFGSM + BIM                                       99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "RFGSM + CW                                        99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "RFGSM + PGD (Linf)                                99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "RFGSM + PGD (L2)                                  99.03                   0.07                      45.83                    3.76                     99.03                   0.04                      37.50                    2.20\n",
            "RFGSM + MIFGSM                                    99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "RFGSM + TPGD                                      99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "RFGSM + EOTPGD                                    99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "RFGSM + APGD                                      99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "RFGSM + DIFGSM (Linf)                             99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "RFGSM + Jitter (Linf)                             99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "PGD (Linf) + GN                                   99.03                   0.07                      87.50                    0.91                     99.03                   0.04                      87.50                    0.47\n",
            "PGD (Linf) + FGSM                                 99.03                   0.07                      87.50                    0.88                     99.03                   0.04                      87.50                    0.45\n",
            "PGD (Linf) + BIM                                  99.03                   0.07                      87.50                    0.91                     99.03                   0.04                      87.50                    0.48\n",
            "PGD (Linf) + CW                                   99.03                   0.07                      87.50                    0.89                     99.03                   0.04                      87.50                    0.45\n",
            "PGD (Linf) + RFGSM                                99.03                   0.07                      87.50                    0.92                     99.03                   0.04                      87.50                    0.47\n",
            "PGD (Linf) + PGD (L2)                             99.03                   0.07                      50.00                    3.56                     99.03                   0.04                      37.50                    2.21\n",
            "PGD (Linf) + MIFGSM                               99.03                   0.07                      87.50                    0.89                     99.03                   0.04                      87.50                    0.48\n",
            "PGD (Linf) + TPGD                                 99.03                   0.07                      87.50                    0.86                     99.03                   0.04                      87.50                    0.46\n",
            "PGD (Linf) + EOTPGD                               99.03                   0.07                      87.50                    0.89                     99.03                   0.04                      87.50                    0.46\n",
            "PGD (Linf) + APGD                                 99.03                   0.07                      87.50                    0.90                     99.03                   0.04                      87.50                    0.48\n",
            "PGD (Linf) + DIFGSM (Linf)                        99.03                   0.07                      87.50                    0.89                     99.03                   0.04                      87.50                    0.47\n",
            "PGD (Linf) + Jitter (Linf)                        99.03                   0.07                      87.50                    0.86                     99.03                   0.04                      87.50                    0.48\n",
            "PGD (L2) + GN                                     99.03                   0.07                      45.83                    5.74                     99.03                   0.04                      37.50                    3.31\n",
            "PGD (L2) + FGSM                                   99.03                   0.07                      45.83                    5.59                     99.03                   0.04                      37.50                    3.06\n",
            "PGD (L2) + BIM                                    99.03                   0.07                      45.83                    5.62                     99.03                   0.04                      37.50                    3.33\n",
            "PGD (L2) + CW                                     99.03                   0.07                      45.83                    5.69                     99.03                   0.04                      37.50                    3.20\n",
            "PGD (L2) + RFGSM                                  99.03                   0.07                      45.83                    5.69                     99.03                   0.04                      37.50                    3.07\n",
            "PGD (L2) + PGD (Linf)                             99.03                   0.07                      45.83                    5.62                     99.03                   0.04                      37.50                    3.19\n",
            "PGD (L2) + MIFGSM                                 99.03                   0.07                      50.00                    5.71                     99.03                   0.04                      37.50                    3.30\n",
            "PGD (L2) + TPGD                                   99.03                   0.07                      50.00                    5.69                     99.03                   0.04                      37.50                    3.23\n",
            "PGD (L2) + EOTPGD                                 99.03                   0.07                      45.83                    5.68                     99.03                   0.04                      37.50                    3.28\n",
            "PGD (L2) + APGD                                   99.03                   0.07                      45.83                    5.60                     99.03                   0.04                      37.50                    3.23\n",
            "PGD (L2) + DIFGSM (Linf)                          99.03                   0.07                      45.83                    5.48                     99.03                   0.04                      37.50                    3.34\n",
            "PGD (L2) + Jitter (Linf)                          99.03                   0.07                      45.83                    5.57                     99.03                   0.04                      37.50                    3.05\n",
            "MIFGSM + GN                                       99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "MIFGSM + FGSM                                     99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "MIFGSM + BIM                                      99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "MIFGSM + CW                                       99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "MIFGSM + RFGSM                                    99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "MIFGSM + PGD (Linf)                               99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "MIFGSM + PGD (L2)                                 99.03                   0.07                      45.83                    3.68                     99.03                   0.04                      37.50                    2.24\n",
            "MIFGSM + TPGD                                     99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "MIFGSM + EOTPGD                                   99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "MIFGSM + APGD                                     99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "MIFGSM + DIFGSM (Linf)                            99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "MIFGSM + Jitter (Linf)                            99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.52\n",
            "TPGD + GN                                         99.03                   0.07                      91.67                    0.52                     99.03                   0.04                      87.50                    0.51\n",
            "TPGD + FGSM                                       99.03                   0.07                      87.50                    0.90                     99.03                   0.04                      87.50                    0.46\n",
            "TPGD + BIM                                        99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.52\n",
            "TPGD + CW                                         99.03                   0.07                      91.67                    0.51                     99.03                   0.04                      91.67                    0.28\n",
            "TPGD + RFGSM                                      99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.53\n",
            "TPGD + PGD (Linf)                                 99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.47\n",
            "TPGD + PGD (L2)                                   99.03                   0.07                      45.83                    4.31                     99.03                   0.04                      37.50                    2.51\n",
            "TPGD + MIFGSM                                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.52\n",
            "TPGD + EOTPGD                                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.51\n",
            "TPGD + APGD                                       99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.51\n",
            "TPGD + DIFGSM (Linf)                              99.03                   0.07                      87.50                    0.96                     99.03                   0.04                      87.50                    0.51\n",
            "TPGD + Jitter (Linf)                              99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.50\n",
            "EOTPGD + GN                                       99.03                   0.07                      87.50                    0.97                     99.03                   0.04                      87.50                    0.52\n",
            "EOTPGD + FGSM                                     99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.52\n",
            "EOTPGD + BIM                                      99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.52\n",
            "EOTPGD + CW                                       99.03                   0.07                      87.50                    0.97                     99.03                   0.04                      87.50                    0.50\n",
            "EOTPGD + RFGSM                                    99.03                   0.07                      87.50                    0.97                     99.03                   0.04                      87.50                    0.53\n",
            "EOTPGD + PGD (Linf)                               99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.50\n",
            "EOTPGD + PGD (L2)                                 99.03                   0.07                      45.83                    3.43                     99.03                   0.04                      37.50                    2.13\n",
            "EOTPGD + MIFGSM                                   99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "EOTPGD + TPGD                                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.51\n",
            "EOTPGD + APGD                                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.53\n",
            "EOTPGD + DIFGSM (Linf)                            99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.53\n",
            "EOTPGD + Jitter (Linf)                            99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.51\n",
            "APGD + GN                                         99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "APGD + FGSM                                       99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "APGD + BIM                                        99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "APGD + CW                                         99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "APGD + RFGSM                                      99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "APGD + PGD (Linf)                                 99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "APGD + PGD (L2)                                   99.03                   0.07                      45.83                    3.65                     99.03                   0.04                      37.50                    2.17\n",
            "APGD + MIFGSM                                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "APGD + TPGD                                       99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "APGD + EOTPGD                                     99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "APGD + DIFGSM (Linf)                              99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "APGD + Jitter (Linf)                              99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.52\n",
            "DIFGSM (Linf) + GN                                99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      91.67                    0.37\n",
            "DIFGSM (Linf) + FGSM                              99.03                   0.07                      87.50                    0.75                     99.03                   0.04                      87.50                    0.38\n",
            "DIFGSM (Linf) + BIM                               99.03                   0.07                      87.50                    0.82                     99.03                   0.04                      87.50                    0.50\n",
            "DIFGSM (Linf) + CW                                99.03                   0.07                      91.67                    0.73                     99.03                   0.04                      87.50                    0.49\n",
            "DIFGSM (Linf) + RFGSM                             99.03                   0.07                      87.50                    0.94                     99.03                   0.04                      87.50                    0.39\n",
            "DIFGSM (Linf) + PGD (Linf)                        99.03                   0.07                      87.50                    0.84                     99.03                   0.04                      87.50                    0.40\n",
            "DIFGSM (Linf) + PGD (L2)                          99.03                   0.07                      41.67                    3.62                     99.03                   0.04                      37.50                    2.04\n",
            "DIFGSM (Linf) + MIFGSM                            99.03                   0.07                      87.50                    0.94                     99.03                   0.04                      87.50                    0.34\n",
            "DIFGSM (Linf) + TPGD                              99.03                   0.07                      87.50                    0.77                     99.03                   0.04                      87.50                    0.43\n",
            "DIFGSM (Linf) + EOTPGD                            99.03                   0.07                      87.50                    0.81                     99.03                   0.04                      87.50                    0.44\n",
            "DIFGSM (Linf) + APGD                              99.03                   0.07                      87.50                    0.72                     99.03                   0.04                      87.50                    0.50\n",
            "DIFGSM (Linf) + Jitter (Linf)                     99.03                   0.07                      87.50                    0.72                     99.03                   0.04                      87.50                    0.45\n",
            "Jitter (Linf) + GN                                99.03                   0.07                      87.50                    0.97                     99.03                   0.04                      87.50                    0.50\n",
            "Jitter (Linf) + FGSM                              99.03                   0.07                      87.50                    0.96                     99.03                   0.04                      87.50                    0.50\n",
            "Jitter (Linf) + BIM                               99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.51\n",
            "Jitter (Linf) + CW                                99.03                   0.07                      87.50                    0.97                     99.03                   0.04                      87.50                    0.49\n",
            "Jitter (Linf) + RFGSM                             99.03                   0.07                      87.50                    0.99                     99.03                   0.04                      87.50                    0.50\n",
            "Jitter (Linf) + PGD (Linf)                        99.03                   0.07                      87.50                    0.94                     99.03                   0.04                      87.50                    0.49\n",
            "Jitter (Linf) + PGD (L2)                          99.03                   0.07                      45.83                    3.60                     99.03                   0.04                      37.50                    2.18\n",
            "Jitter (Linf) + MIFGSM                            99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.50\n",
            "Jitter (Linf) + TPGD                              99.03                   0.07                      87.50                    0.93                     99.03                   0.04                      87.50                    0.51\n",
            "Jitter (Linf) + EOTPGD                            99.03                   0.07                      87.50                    0.98                     99.03                   0.04                      87.50                    0.51\n",
            "Jitter (Linf) + APGD                              99.03                   0.07                      87.50                    1.00                     99.03                   0.04                      87.50                    0.51\n",
            "Jitter (Linf) + DIFGSM (Linf)                     99.03                   0.07                      87.50                    0.94                     99.03                   0.04                      87.50                    0.50\n",
            "\n",
            "Top 3 Compound Attacks with Greatest Accuracy Difference for CNN:\n",
            "Attack                      Accuracy Difference (%)\n",
            "------------------------  -------------------------\n",
            "FGSM + PGD (L2)                               57.36\n",
            "DIFGSM (Linf) + PGD (L2)                      57.36\n",
            "GN + PGD (L2)                                 53.20\n",
            "\n",
            "Top 3 Compound Attacks with Greatest Accuracy Difference for HNN:\n",
            "Attack             Accuracy Difference (%)\n",
            "---------------  -------------------------\n",
            "GN + PGD (L2)                        61.53\n",
            "FGSM + PGD (L2)                      61.53\n",
            "BIM + PGD (L2)                       61.53\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 1200x600 with 2 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    }
  ],
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "gpuType": "L4",
      "machine_shape": "hm",
      "provenance": [],
      "authorship_tag": "ABX9TyOM4PrMiDAqpDsm2wLa3jPH",
      "include_colab_link": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}