docs/source/notebooks/aijack_gradient_inversion_attack.ipynb

{
 "cells": [
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {
    "id": "k0rUERH-551c"
   },
   "source": [
    "# Gradient-based Model Inversion Attack against Federated Learning\n",
    "\n",
    "In this tutorial, we will experiment gradient-based model inversion attack, which allows the malicious server of Federated Learning to reconstruct the private local dataset via shared gradients. You can implement five popular gradient-based model inversion attacks with AIJack. These methods reconstruct the private images by minimizing the distance between the fake gradients and the received gradients. Each method has its own strategy, such as the distance metric and regularization terms.\n",
    "\n",
    "One example is as follows. Since the server already knows the parameters of the global model $w_{t - 1}$, the server can estimate the private training sample $(X, Y)$ with the following optimization.\n",
    "\n",
    "$$\n",
    "X' \\leftarrow X' - \\lambda \\nabla_{X'} D\n",
    "$$\n",
    "\n",
    "$$\n",
    "Y' \\leftarrow Y' - \\lambda \\nabla_{Y'} D\n",
    "$$\n",
    "\n",
    ", where $D$ is the loss function calculated as follows:\n",
    "\n",
    "$$\n",
    "D = || \\nabla \\mathcal{l}(w_{t - 1}, X, Y) - \\nabla \\mathcal{l}(w_{t - 1}, X', Y') ||_{2}\n",
    "$$\n",
    "\n",
    "In other words, this attack tries to reconstruct the private training data by optimizing the fake data to generate gradients close enough to the received gradients from the client."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "background_save": true
    },
    "id": "2b1JISUQ5MXM"
   },
   "outputs": [],
   "source": [
    "import cv2\n",
    "import copy\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import torchvision\n",
    "import torchvision.transforms as transforms\n",
    "from numpy import e\n",
    "from matplotlib import pyplot as plt\n",
    "import torch.optim as optim\n",
    "from tqdm.notebook import tqdm\n",
    "\n",
    "from aijack.collaborative.fedavg import FedAVGAPI, FedAVGClient, FedAVGServer\n",
    "from aijack.attack.inversion import GradientInversionAttackServerManager\n",
    "from torch.utils.data import DataLoader, TensorDataset\n",
    "from aijack.utils import NumpyDataset\n",
    "\n",
    "import warnings\n",
    "\n",
    "warnings.filterwarnings(\"ignore\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "background_save": true
    },
    "id": "CrFIlyBs5Se2"
   },
   "outputs": [],
   "source": [
    "class LeNet(nn.Module):\n",
    "    def __init__(self, channel=3, hideen=768, num_classes=10):\n",
    "        super(LeNet, self).__init__()\n",
    "        act = nn.Sigmoid\n",
    "        self.body = nn.Sequential(\n",
    "            nn.Conv2d(channel, 12, kernel_size=5, padding=5 // 2, stride=2),\n",
    "            nn.BatchNorm2d(12),\n",
    "            act(),\n",
    "            nn.Conv2d(12, 12, kernel_size=5, padding=5 // 2, stride=2),\n",
    "            nn.BatchNorm2d(12),\n",
    "            act(),\n",
    "            nn.Conv2d(12, 12, kernel_size=5, padding=5 // 2, stride=1),\n",
    "            nn.BatchNorm2d(12),\n",
    "            act(),\n",
    "        )\n",
    "        self.fc = nn.Sequential(nn.Linear(hideen, num_classes))\n",
    "\n",
    "    def forward(self, x):\n",
    "        out = self.body(x)\n",
    "        out = out.view(out.size(0), -1)\n",
    "        out = self.fc(out)\n",
    "        return out"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "background_save": true
    },
    "id": "7jfBlFUS5Uk3"
   },
   "outputs": [],
   "source": [
    "def prepare_dataloader(path=\"MNIST/.\", batch_size=64, shuffle=True):\n",
    "    at_t_dataset_train = torchvision.datasets.MNIST(\n",
    "        root=path, train=True, download=True\n",
    "    )\n",
    "\n",
    "    transform = transforms.Compose(\n",
    "        [transforms.ToTensor(), transforms.Normalize((0.5,), (0.5,))]\n",
    "    )\n",
    "\n",
    "    dataset = NumpyDataset(\n",
    "        at_t_dataset_train.train_data.numpy(),\n",
    "        at_t_dataset_train.train_labels.numpy(),\n",
    "        transform=transform,\n",
    "    )\n",
    "\n",
    "    dataloader = torch.utils.data.DataLoader(\n",
    "        dataset, batch_size=batch_size, shuffle=shuffle, num_workers=0\n",
    "    )\n",
    "    return dataloader"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "background_save": true
    },
    "id": "W1SuaQLL7aTg"
   },
   "outputs": [],
   "source": [
    "torch.manual_seed(7777)\n",
    "\n",
    "shape_img = (28, 28)\n",
    "num_classes = 10\n",
    "channel = 1\n",
    "hidden = 588\n",
    "\n",
    "num_seeds = 5"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "Zpt1sJxU9lVi"
   },
   "source": [
    "## Reconstruct Single Data\n",
    "\n",
    "First, we try to recover the following private image from the received gradients with the batch size of 1."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "background_save": true,
     "referenced_widgets": [
      "3569bb4d4af64a7290063559ef31f680",
      "909e2f5c8bf94c409941c22bcc8a2945",
      "88ccdc52d9e94985b3bf9d38f0d51663",
      "36f26d71566f4a8285c1ee2183fc3994"
     ]
    },
    "id": "z-dkdI2E5gE_",
    "outputId": "75a6aef5-53c3-4b09-a66a-2f9ba0f69b8d"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading http://yann.lecun.com/exdb/mnist/train-images-idx3-ubyte.gz\n",
      "Downloading http://yann.lecun.com/exdb/mnist/train-images-idx3-ubyte.gz to MNIST/./MNIST/raw/train-images-idx3-ubyte.gz\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "3569bb4d4af64a7290063559ef31f680",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/9912422 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Extracting MNIST/./MNIST/raw/train-images-idx3-ubyte.gz to MNIST/./MNIST/raw\n",
      "\n",
      "Downloading http://yann.lecun.com/exdb/mnist/train-labels-idx1-ubyte.gz\n",
      "Downloading http://yann.lecun.com/exdb/mnist/train-labels-idx1-ubyte.gz to MNIST/./MNIST/raw/train-labels-idx1-ubyte.gz\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "909e2f5c8bf94c409941c22bcc8a2945",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/28881 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Extracting MNIST/./MNIST/raw/train-labels-idx1-ubyte.gz to MNIST/./MNIST/raw\n",
      "\n",
      "Downloading http://yann.lecun.com/exdb/mnist/t10k-images-idx3-ubyte.gz\n",
      "Downloading http://yann.lecun.com/exdb/mnist/t10k-images-idx3-ubyte.gz to MNIST/./MNIST/raw/t10k-images-idx3-ubyte.gz\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "88ccdc52d9e94985b3bf9d38f0d51663",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/1648877 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Extracting MNIST/./MNIST/raw/t10k-images-idx3-ubyte.gz to MNIST/./MNIST/raw\n",
      "\n",
      "Downloading http://yann.lecun.com/exdb/mnist/t10k-labels-idx1-ubyte.gz\n",
      "Downloading http://yann.lecun.com/exdb/mnist/t10k-labels-idx1-ubyte.gz to MNIST/./MNIST/raw/t10k-labels-idx1-ubyte.gz\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "36f26d71566f4a8285c1ee2183fc3994",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/4542 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Extracting MNIST/./MNIST/raw/t10k-labels-idx1-ubyte.gz to MNIST/./MNIST/raw\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 72x72 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "device = torch.device(\"cuda:0\") if torch.cuda.is_available() else \"cpu\"\n",
    "dataloader = prepare_dataloader()\n",
    "for data in dataloader:\n",
    "    xs, ys = data[0], data[1]\n",
    "    break\n",
    "\n",
    "x = xs[:1]\n",
    "y = ys[:1]\n",
    "\n",
    "fig = plt.figure(figsize=(1, 1))\n",
    "plt.axis(\"off\")\n",
    "plt.imshow(x.detach().numpy()[0][0], cmap=\"gray\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "background_save": true
    },
    "id": "UA3YUHqi6nq2"
   },
   "outputs": [],
   "source": [
    "criterion = nn.CrossEntropyLoss()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "zsetq0Lc-NVd"
   },
   "source": [
    "### DLG\n",
    "\n",
    "You can convert the normal server to the malicious attacker with `GradientInversionAttackServerManager`.\n",
    "\n",
    "https://dlg.mit.edu/assets/NeurIPS19_deep_leakage_from_gradients.pdf\n",
    "\n",
    "- distance metric = L2 norm\n",
    "- optimize labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 161
    },
    "id": "4BAe6aEu5odr",
    "outputId": "027167b7-1a67-4bce-be42-d4b135d72767"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "communication 0, epoch 0: client-1 2.285383462905884\n",
      "iter=80: loss did not improve in the last 50 rounds.\n",
      "iter=73: loss did not improve in the last 50 rounds.\n",
      "iter=70: loss did not improve in the last 50 rounds.\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAWAAAABJCAYAAAADtfwGAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAO+ElEQVR4nO3d22/T9v/H8Zft2DnazrlpE3oga0sP2gqlDCQQEwhN7Ha7mXa5f2wXuxvijm0S0jhoMCZNonTQru1S0kMKpCVNmrOdxN8LFP9o+cCPL98mrsv7IXHB1qLmqU/fSZyPbc4wDBBCCOk+3uofgBBCPlY0gAkhxCI0gAkhxCI0gAkhxCI0gAkhxCI0gAkhxCKOd/1PjuM+qj1qhmFw7/N11IWNurBRFzbqQq+ACSHEMjSACSHEIjSACSHEIjSACSHEIjSACSHEIjSACSHEIu/chmY1h8MBQRDQ29sLRVHQ39+PWCwGp9MJl8uFzc1NPHnyBLu7u9jY2ECj0bD6R+4K6sJGXdioC9th6HJoBzDHcRBFEU6nEyMjIxgYGMClS5dw+vRp+P1+BINBPHjwAD/88ANWV1fx/Pnzj2LhUBc26sJGXdgOS5dDM4B5nofT6YQkSQiFQnC73ejv74eqqpiYmEAsFsPAwAAURYHL5QLP81AUBaOjowAAp9MJTdPQarUsfiQHi7qwURc26sJ2WLscmgEsiiICgQD8fj9mZmbQ09ODK1euIB6PIxKJwOfzQRAE8DwPjuNgGAZ6e3tx6dIlyLKMGzduoFarQdM0HKWLzFMXNurCRl3YDmsXywewy+WCLMsIBoMYGxtDIBDA2NgYgsEgotEo/H4/PB4PJEkCx/3fmXyGYUCSJKiqikAggGAwiGaziZcvXx6Jt1Af0sUwDOpC64W6vOawd7F8AIfDYZw8eRLj4+P4/vvv4ff7IYoiBEGAJEkQBGHPgHmdx+PBsWPHkM/ncfLkSayvr+Ovv/5CqVSy4qEcKOrCRl3YqAvbYe9i2QDmed48zpJIJJBIJBCNRiHL8p6va78dYOE4Dg6HAx6PB729vdA0DQ6H5c8p/5P37fIu1IWNurBRF7ZudLGsssvlgs/nw+TkJL799luEw2G43e49X/P6W4L9f399KEciEXz11VdYWlrC/fv3kc/nO/vDd9D7dAH2PjFRl1f2v5KhLq9Qlw/r0v57J7tYOoCDwSBCoRBisRhUVd1zABwAWq0WDMNArVaDrut7jnE6nU5wHAdBECCKIkKhEILBIARBsOohHYj36WIYBlqtFnWh9UJd3sIuXSwbwOPj47h69SrGxsbQ09MDp9MJnuf3DJlqtYp6vY7ffvsNy8vL0DQNmqZhcnISp06dgqIo6OnpAc/zcLlc5r9hZ9SFjbqwURc2u3Tp+gDmOA4cxyEcDmNkZATxeBxutxuCIOx5O9R+hVcul5FOp/H48WNzG4jP50MymYQoikdmqwx1YaMubNSFzW5duj6Aw+EwVFXF2NgYpqamoCgKeJ5Ho9FAsVg03xZUq1Xcu3cPmUwGN2/exPLyMprNJprNJnw+H3p7e5FMJtHf349qtYpUKoV0Og1N07r9kA4EdWGjLmyhUAh+v/+NLrquo1QqfbRd7LZeuj6AvV4vwuEwenp6EI/HzWMqjUYD5XIZjUYDhmGgWCxiYWEBqVQKS0tLWF1dBfDqGe758+fIZrMIh8MwDAP1eh1bW1u23rtIXdgOskskEjkyXXw+H7NLs9mk9WKj36OuDmCO43DixAmcPXsWJ06cgMPhQKlUQjabxerqKm7cuIFyuYxWqwVd17G8vIxCoYCdnR0AQF9fHyKRCGZmZnDhwgX4/X6USiWsra3h559/xubmpi33LlIXtoPoEo1GmV1+/fVXbGxsfLRdaL0cjvXS9QE8MDCAs2fPYnBwEIIgoFqtYn19HbOzs/jxxx+xs7PDPO7CcRwikQhGRkYwMTGBqakp6LqOYrGIZ8+e4Y8//kA2m0WtVuvmQzoQ1IXtILoMDw8zu9y7d++j7kLr5c3vtWK9dG0Ai6IIURShKAqCwaC5J69YLCKVSiGTyZgv7zmOgyRJGBsbg6qqkGUZLpcL4+PjGBoawtjYGHieRy6Xw99//435+XlUKhXz7YWdUBc26sJGXdjs2qVrA1iSJLhcLiiKYl78AgAKhQIWFhaQTqfNvXg8z8PtduPcuXNIJpNIJBLw+/345JNPkEgkzE86s9ks7t69i5WVFZRKJVt+cEBd2KgLG3Vhs2uXrgzg1zc0ezweeL1eSJIE4NV+vGazuWcD9MjICCKRCD799FMkEgmEw2F4PB7zknDb29vIZrOYm5vDP//8gxcvXtjyQwPqwkZd2DrV5dGjR9TFovViySvgYDBonubXarXMB+fxeBAMBvHdd99hcHAQMzMzCIVC5teWSiUUCgXMzs7izp07WFxcxO3bt6HrOprNZrceyoGiLmzUhY26sNm1S9cG8Oun/7VaLfO0QFmWMTw8jEAgAEmSoCgKBgcHEYvF4Ha74XA4UKvV0Gg0kE6nkclksLCwgKdPnyKbzdp60QDU5W2oC9tBdnny5Al1sXi9dG0At5+JNE1DrVaDKIqQJAnDw8NIJBJoNpvQNA08z8Pr9cLhcJhnomxtbSGfz+Onn37CzZs3zbcIjUbD1osGoC5vQ13YqAubXbt0bQA3Gg3zLJ1cLgdZliFJkvnpZftCGADM8605jjM3ludyORSLRVSrVVSrVVQqlW796B1FXdioCxt1YbNrl64MYMMwUC6XUavVMDc3h2g0iomJCUxPT5ufOLb/7P++ZrOJlZUVLC4uQtM0JBIJNBoNZLNZ222V2Y+6sH1oFwDUhdaLrbp07ZJH7bcI+Xwem5ub2Nrawu7uLqrVqnl+9v9HlmXEYjEEAgG43W7zk047oy5sH9Jl/3+jLmzUhc2KLl09E84wDKRSKWiaho2NDaysrGBoaAhnzpyB2+2G1+vdc8m49hXpz5w5g8nJSdTrdei6jlu3bkFVVWQyGczPz9v+Dq7UhY26sFEXNjt26doAbt8ipF6vo1AoYHt7G5lMBoqiQNd1iKL4xve0H3goFDK3i3Ach7W1NcRiMZTLZebbUDuhLmzUhY26sNm1S1cGMM/zGBwcRDAYxJUrV/D5559DURSoqgpVVc0b5e2/Wv3Ozg50XTdvnud2u+F0OuH1etHX14dcLmfrhXOQXSRJoi7UhX6PbNalawM4GAzi2LFjOHXqFC5fvmx+Ogm8eePN9sHxUqmEarUKj8cDh8MBh8Nhns2iqio8Ho/tFw51eRN1YaMubHbu0tEB3A6jKAouX76MqakpjI+PQ5KkPVtB2ltEdF1HpVLB0tIScrkc7ty5g2w2i1AoBJ/Ph4sXL2J6ehqBQACnT59GoVCw5a1TqAsbz/MIhUKQZZm6vKbT6yWfz9u2i93XS8cHsN/vRzgcxpkzZ/DFF1+YZ58Ae+9C2r5GZ7FYxPz8PNbW1nDt2jWk02nE43EEAgFEo1GcOHECiqJgdHQUCwsLtl041OVNPM9DVVXqsg+tF7Z3rZf9r1wPa5eODOD2TexUVcWXX36JZDKJZDIJSZJQqVSQz+dRKBSQy+UgSRI8Hg9KpRJSqRRevnyJBw8emGenGMaru5aWSiVUKhXUajXzYLndPrWlLmzUhY26sB2lLh0bwLIso6enB19//TVmZmbgdDohCAJyuRyy2ay5+VmWZUSjUTx//hy3b9/G1tYWHj9+jHK5bB63qVarMAwDpVIJ5XLZPF5jtys3dbOLnTbX03pho/XCdpTWS0cGsCRJiMfjSCQS8Pl85qmA7QtezM3NYWNjA6lUynwmKxaL2NnZQaPRQCKRMN8utFotjI6OIhqNIplMQlVVZLNZZDIZbGxs2OrZu5td7PQLReuFjdYL21FaLx0ZwG63G5999hmGhoYQDAYhiiJ0XYeu6/jzzz9x7do180ECr45huVwuhMNhKIqC6elpiKKIlZUV1Ot1fPPNNzh58iSGhobQ19eH+fl5/PLLL1hcXLTVRUSoCxt1YaMubEepS0cGsCAI5h689gcFmqahWq2ax2aKxSJ0XQfHceB5HpIkoa+vD6FQCKOjo+a1PXVdx8DAAEKhEGq1GlZXV7G2tob19XXkcjlbPXNTFzbqwvYhXURRpC42Wi8dOwRx7NgxDA4Owu12wzAMFAoF5PN5ZDIZpNNp86W9IAhwuVyIx+O4evUq4vE4zp8/D1mWzSvZ+3w+OBwO/P7775idncX9+/dx9+5daJpmq2du6sJGXdg+pEsikfivuty5c8d21wI+SuulYx/Ceb1eeL1eCIIA4NU2kGazCVEU4fV6zWem9lXqh4aGkEgkEIvF4Pf74fV6zYsrty+qsb6+jqdPn+LFixeoVCq2Op4HUJe3oS5s3ejSvliNnRyl9dKRAdx+G5RIJOB0Ovdci3NoaAjnz583byEyODiIc+fOIRAIYGRkBE6n0zwDpb015Pr163j8+DEePXqEf//9F9Vq1VbP2G3UhY26sO3vAoC64Gh16cgA5jjOPBWwfWsQQRAgCAJCoRD6+/vhcrngdDpx/PhxJJNJyLKMQCAAjuPMA+rb29soFApYXV1FKpXC5uYmtre3O/EjdwV1YaMubPu7AKAuOFpdOjKADcNApVJBpVJBs9kEx3EIhULmgfMLFy6YVy9yu93w+/3geR7NZhOVSgXz8/PY2trC9evXkU6nsbGxgd3dXdtfvZ+6sFEXtlardWBdnj59ikwmcyS6HKX10pEB3H6g5XIZmqbB6XSaF7nwer0AYL7NbP9pf8/u7i6ePXuGTCaDhw8fYnl5GY1Gw3bHqVioCxt1YWsPYOqy11FaL9y7tllwHPdBezDaz0KxWAwXL15EPB7H8PAwIpEIBEEAz/PI5/PY3t4278e0vb2Nhw8fIp/PY2lpCYVCAalUytws3Q2GYbzXpY+oCxt1YaMubNSlQ6+AK5UKZmdn4ff70dPTg1qthkgkAlVVIYoiBEFAqVRCNps1byGyvr6OW7duIZfLYX19HfV6vRM/mqWoCxt1YaMubEepS0deAbe3f7hcLhw/fhyqqqK3txeyLJvHZsrlMorFIur1OkqlEorFIlZXV1Gr1VAsFi35dLbTz9zUhY26sFEXtqPUpSMD2K46vXDsirqwURc26sLG6mK/i4ASQsgRQQOYEEIsQgOYEEIsQgOYEEIsQgOYEEIsQgOYEEIsQgOYEEIs8s59wIQQQjqHXgETQohFaAATQohFaAATQohFaAATQohFaAATQohFaAATQohF/gMaq30+ziHwrQAAAABJRU5ErkJggg==",
      "text/plain": [
       "<Figure size 360x144 with 5 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "manager = GradientInversionAttackServerManager(\n",
    "    (1, 28, 28),\n",
    "    num_trial_per_communication=5,\n",
    "    log_interval=0,\n",
    "    num_iteration=100,\n",
    "    distancename=\"l2\",\n",
    "    device=device,\n",
    "    lr=1.0,\n",
    ")\n",
    "DLGFedAVGServer = manager.attach(FedAVGServer)\n",
    "\n",
    "client = FedAVGClient(\n",
    "    LeNet(channel=channel, hideen=hidden, num_classes=num_classes).to(device),\n",
    "    lr=1.0,\n",
    "    device=device,\n",
    ")\n",
    "server = DLGFedAVGServer(\n",
    "    [client],\n",
    "    LeNet(channel=channel, hideen=hidden, num_classes=num_classes).to(device),\n",
    "    lr=1.0,\n",
    "    device=device,\n",
    ")\n",
    "\n",
    "local_dataloaders = [DataLoader(TensorDataset(x, y))]\n",
    "local_optimizers = [optim.SGD(client.parameters(), lr=1.0)]\n",
    "\n",
    "api = FedAVGAPI(\n",
    "    server,\n",
    "    [client],\n",
    "    criterion,\n",
    "    local_optimizers,\n",
    "    local_dataloaders,\n",
    "    num_communication=1,\n",
    "    local_epoch=1,\n",
    "    use_gradients=True,\n",
    "    device=device,\n",
    ")\n",
    "\n",
    "api.run()\n",
    "\n",
    "fig = plt.figure(figsize=(5, 2))\n",
    "for s, result in enumerate(server.attack_results[0]):\n",
    "    ax = fig.add_subplot(1, len(server.attack_results[0]), s + 1)\n",
    "    ax.imshow(result[0].cpu().detach().numpy()[0][0], cmap=\"gray\")\n",
    "    ax.axis(\"off\")\n",
    "plt.tight_layout()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "G1U9EU7UAsju"
   },
   "source": [
    "### GS Attack\n",
    "\n",
    "https://arxiv.org/abs/2003.14053\n",
    "\n",
    "- distance metric = cosine similarity\n",
    "- optimize labels\n",
    "- regularization: total-variance"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 178
    },
    "id": "PJAu_poY6q7L",
    "outputId": "30d1c8bc-3f36-44e8-aee7-486d6609254c"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "communication 0, epoch 0: client-1 2.371312141418457\n",
      "iter=89: loss did not improve in the last 50 rounds.\n",
      "iter=72: loss did not improve in the last 50 rounds.\n",
      "iter=71: loss did not improve in the last 50 rounds.\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 360x144 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "manager = GradientInversionAttackServerManager(\n",
    "    (1, 28, 28),\n",
    "    num_trial_per_communication=5,\n",
    "    log_interval=0,\n",
    "    num_iteration=100,\n",
    "    tv_reg_coef=0.01,\n",
    "    distancename=\"cossim\",\n",
    "    device=device,\n",
    "    lr=1.0,\n",
    ")\n",
    "GSFedAVGServer = manager.attach(FedAVGServer)\n",
    "\n",
    "client = FedAVGClient(\n",
    "    LeNet(channel=channel, hideen=hidden, num_classes=num_classes).to(device),\n",
    "    lr=1.0,\n",
    "    device=device,\n",
    ")\n",
    "server = GSFedAVGServer(\n",
    "    [client],\n",
    "    LeNet(channel=channel, hideen=hidden, num_classes=num_classes).to(device),\n",
    "    lr=1.0,\n",
    "    device=device,\n",
    ")\n",
    "\n",
    "local_dataloaders = [DataLoader(TensorDataset(x, y))]\n",
    "local_optimizers = [optim.SGD(client.parameters(), lr=1.0)]\n",
    "\n",
    "api = FedAVGAPI(\n",
    "    server,\n",
    "    [client],\n",
    "    criterion,\n",
    "    local_optimizers,\n",
    "    local_dataloaders,\n",
    "    num_communication=1,\n",
    "    local_epoch=1,\n",
    "    use_gradients=True,\n",
    "    device=device,\n",
    ")\n",
    "\n",
    "api.run()\n",
    "\n",
    "fig = plt.figure(figsize=(5, 2))\n",
    "for s, result in enumerate(server.attack_results[0]):\n",
    "    ax = fig.add_subplot(1, len(server.attack_results[0]), s + 1)\n",
    "    ax.imshow(result[0].cpu().detach().numpy()[0][0], cmap=\"gray\")\n",
    "    ax.axis(\"off\")\n",
    "plt.tight_layout()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "dHNW787NAhFR"
   },
   "source": [
    "### iDLG Attack\n",
    "\n",
    "https://arxiv.org/abs/2001.02610\n",
    "\n",
    "- distance metric = L2 norm\n",
    "- analytically estimate a label from the gradients"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 108
    },
    "id": "bv0qW14888Kh",
    "outputId": "1abf4cd3-ca40-4c82-c0e3-3d1c67c2289b"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "communication 0, epoch 0: client-1 2.371312141418457\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 360x144 with 5 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "manager = GradientInversionAttackServerManager(\n",
    "    (1, 28, 28),\n",
    "    num_trial_per_communication=5,\n",
    "    log_interval=0,\n",
    "    num_iteration=1000,\n",
    "    optimizer_class=torch.optim.SGD,\n",
    "    distancename=\"l2\",\n",
    "    optimize_label=False,\n",
    "    device=device,\n",
    "    lr=1.0,\n",
    ")\n",
    "iDLGFedAVGServer = manager.attach(FedAVGServer)\n",
    "\n",
    "client = FedAVGClient(\n",
    "    LeNet(channel=channel, hideen=hidden, num_classes=num_classes).to(device),\n",
    "    lr=1.0,\n",
    "    device=device,\n",
    ")\n",
    "server = iDLGFedAVGServer(\n",
    "    [client],\n",
    "    LeNet(channel=channel, hideen=hidden, num_classes=num_classes).to(device),\n",
    "    lr=1.0,\n",
    "    device=device,\n",
    ")\n",
    "\n",
    "local_dataloaders = [DataLoader(TensorDataset(x, y))]\n",
    "local_optimizers = [optim.SGD(client.parameters(), lr=1.0)]\n",
    "\n",
    "api = FedAVGAPI(\n",
    "    server,\n",
    "    [client],\n",
    "    criterion,\n",
    "    local_optimizers,\n",
    "    local_dataloaders,\n",
    "    num_communication=1,\n",
    "    local_epoch=1,\n",
    "    use_gradients=True,\n",
    "    device=device,\n",
    ")\n",
    "\n",
    "api.run()\n",
    "\n",
    "fig = plt.figure(figsize=(5, 2))\n",
    "for s, result in enumerate(server.attack_results[0]):\n",
    "    ax = fig.add_subplot(1, len(server.attack_results[0]), s + 1)\n",
    "    ax.imshow(result[0].cpu().detach().numpy()[0][0], cmap=\"gray\")\n",
    "    ax.axis(\"off\")\n",
    "plt.tight_layout()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "-CTIH-k_AJ3x"
   },
   "source": [
    "### CPL Attack\n",
    "\n",
    "https://arxiv.org/abs/2004.10397\n",
    "\n",
    "- distance metric = L2 norm\n",
    "- analytically estimate an label from the gradients\n",
    "- regularization: label-matching"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 108
    },
    "id": "Lz-UJfvR9MDZ",
    "outputId": "d343c25b-4319-4586-9bb4-fab2deaa2069"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "communication 0, epoch 0: client-1 2.371312141418457\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAWAAAABJCAYAAAADtfwGAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nO29WXNbV5bn+wdAgpgJgAQBUtRsW4Mlz5WV5eqsiqyqfugaoiuqoyP6W/WXqaqH7ujozi6nLdtppQfJg1K2JM4TSEwEMeM+qH+L6xxRFOG+cfPhckcoJJHAOXuvvcb/WnvtyHg81vk4H+fjfJyP/+9H9I89gfNxPs7H+fj/6zhXwOfjfJyP8/FHGucK+Hycj/NxPv5I41wBn4/zcT7Oxx9pnCvg83E+zsf5+CONcwV8Ps7H+Tgff6Qxddov79y5M45EIpqZmVEkEpEkjcdjRSIRxWIx+1wkElE0GtVwOFS329V4PBblbdHocx0fi8UUiUQ0GAw0Go00Go00Ho8Vi8U0PT2tSCSiqakpjcdjdbvdF94ViUTsmbxvNBppMBhoampK8Xhckuz5jHg8HvhsPB7XzMyM+v2+Op2OpqenlcvlFI1G9c///M+RsxDtnXfeGUciESUSCU1NTWkwGLyULpI0Go2MLsxtampK0WjU6DMYDDQYDGzdni48s9/vG914Bu/ge7xjMBgoFospmUxqPB7bd8OD78TjcSWTSdvDVqultbU1DQYDtVqtM9Hl8uXLxi/RaNSeHY1GFYvFbG6xWEyxWEzD4VCdTidAq/DawnSJRCKKx+OKRCL2HX7P3zzfP9c/KxaLGU97noIHY7GYut2uer2e8vm85ubm1G63tbe3p0KhoLt372pmZkb/9b/+1zPR5cqVK0aXWCxm++D3n7VJz/ml1+sF9ovP+YEcDYdDRaNRo9nU1FTg9/59yJJ/3nA4tGd4OfL7kUwmjddHo5FSqZTS6bT6/b7a7bYGg4HN+bvvvjszv0SjUZOjfr+v4XBoe+D5ZWpqSsPhUEdHR/Kls+yb1y/MHfryO+gCz/EZzyOebuPx2OaTSCTss+Px2NbK87vdrgaDgdLptDKZjNHF793Dhw9foMupCjibzUoKKrHRaKSjoyO1Wi0lEgkT8OFwGCAMmw6BIABE6PV66na7AQLy+ZmZGRPOwWCgRCIREDoI2+l01Ol0jKk9IXlPq9VSr9dTOp1WMplUp9PRwcGBEomEcrmcvfck5fQqusDw09PTGo/HOjo60uHhoWZmZpRIJIwuzC0SiZz4Hhg/Ho+r1+up0+kE1sq64vG4RqORDg8PNRwOA+vkc5FIxOgyNTWl6elpSc+FTJISiYRisZjRJZPJKJFIaDAYqN1uK5lManFxUbFYTG+99daZaSJJmUxGGKZYLGbCCl1YIwzJfL1h8j9HEBH8fr+vWCxm+++N+3A4tPexp97IQcNOp6PRaBRQVuHfI8CSVK/X1Wq1bE7tdlsPHz4MGL5XjWQyaXuMozIej9XpdNTtdjU9PR3gl/F4bEYdxewVkTdWKC0vR/DEYDBQJBIJKDZPF0mmwI+OjgL7EjaE7XZbw+FQMzMzmp6e1t7enp49e6bZ2VlVKhUNBgPVarVJ2EW5XE6STLZxwDqdjsnRzMyMRqORORB8ptfr2Xf9HsdiMSUSCfV6PfV6PUWjUZMBaIYcwQsYAOiB49Pr9XR0dKTRaKR4PG77Ih0beQx1IpEwxXtwcKB4PK58Ph+Qy5PGqQqYSXmPBc8Gq+kXdpoS8wLD/8OT8kyBBYQRIL4XPj+XwWDwgrBJMsucSCTs/ygnhNl7H2cZWFTvmWMx8cg9XV41Ttsg762gyHkf7wxbcN7rBfik5zG36elpDQYDE+R4PK5UKqW5ubmAR/+qAf2hrecH5ut54GW0wWvG6IcNLOtgbsPh8MR5egMVficKyfORpICXjPIiUkKw9/f3z0wTKShH3gj7SMXvF0qYObGH3jv0SvdlvAP9ws/yv+NnyPLLZLjX69n7p6en1ev11Gg0THl65T0pXbwhDdPAe6X8P+wB87d3WE6KGMJRuZcjTzPPW+EoAjqxl/wMXuv3++r1euZsomNeSoPTCLS3t6doNKpcLqepqSnzDmKxmDKZjAaDgQ4ODixcxmuNRqNKpVKKxWI6OjrSYDDQzMyMpqam1G63zTIlEgmzRFNTU0qn0+ZJYm0TiUTAysTj8UCYkclkNBwO7ZmpVMosNh5RPB5Xv983C43Xvre3p5mZGRWLxYkUzcHBgSKRiNLptKLRqHq9nhmAVCql4XCoWq0WoMvR0ZEikYgymYwk2XfS6bSmp6cD3nwqlQrQhf+3220zRJ7e09PTBqv0+32ji4dzCKGw+tFoVMlkUpLUbrdtbv1+X8+ePVM2m9X09PQLCuxVdEGw4/G42u22ea2pVMq8JOaL1x2NRo0uzA8hIiIgDPR0IfKALnhDCAHeT7/fV7fbVTQaVTabtRB/PB6bYW42m+r3+ybIeI/SsYGqVquanp5WNpudSNFAl3Q6HZAj9mAwGKherxu/4CSwJ7FYLEAXBL3b7RrMhAxgKIbDYYAuPkz2HnKv17P34FkSxUQiEbVarYCy5xmRSESzs7MajUZaXV01z3NSukQiEaVSKU1NTanb7ZqThxzV63Uz6kB56A4M5Hg8Noik0+nYPhJVoJPwYtEv09PTmp6eNtmKx+PGL0dHR5KkdDotSfZMlKmXI2jL3DOZjDqdjn766Sclk0mVSqWXytGpSbh2u20KE6ZgYjDu0dGRWUcs+nA4tMUQ4jB5Hz54nM97OP1+3xiF8IF34GWCwaAkvIdF+ADWA37U6XQsBGF9Hg866+B7zKHb7dpzvMJlUwifMV68nzUTrqMUUCRh/DMcVhGCMhDC0WhkBg/aYpF5jySbB88gNDs4OFCtVjOjcNZxdHRkAgD9wyEcdCOC8nTxYaCkF+gCv8BrDE8X1uQhMd4DXaCdj+C63a4JJvTm3xgV9n3S4/tHR0fmiLAe6Mp8oRsjvG/QBYHHiEjPeQ5oA1mTZAYZOfLwB/ICfyUSCeNd/wzCcGgBj6HgpOeKtNVqGa0moQv8ACSJYmOPPF2AItANyDUGAbrgdMAvQDCMsCx6iMOvESUdfg86i+/AH8wdXt/f31etVnshL+VH5DRm+rM/+7OxZwYPNGN1w0k1wkEslE9QoVTC+CWDhaBEwBI98aTjUOKkhAO/D3+HUBMhwOPudruq1WoajUb68ccfz2S+f/WrXwXogkeTSqXMSwszOhuFN+JDLO9FE7awPj8QWp8U9TTx4VEYBvGJGI8hQjufxMBjgIHv3bt3JrrcvXvX6IKi8XTxmLiHbBBmHxYz/H75sO8kunjMmFDVh/Osya8d2oSTXh5e8hCTp+Xq6uqZ6HLr1i2jCx5Yv99XNps1Dxglg4GBLnhgHntEqaBEUJzwEzgpCSuMcRiS4j1TU1OWd/EJLEkBuebnYSUb/s5Z5Qh+QUHydyKR0MzMTEB+vGEk73KSbvAO30n84qPCk3jKwzVhXJ3vhmE9/38gh36/r1arFUjy//73v58sCQe4vrOzE/BA2ZSpqSklk0n1ej01m82AVYQweHNYO6wx1hdF6EMmv8lsyPT0dOAZXsH4MKzRaFiWlpAEZR6Px9VqtXR0dKREIqFisahqtarV1VXblLOM+fl5DYdD7e/vn1hdAMN3u10dHh4GPDPWBP7Y7XYtGcmafObVW1Z+j9L2sI6ni2cGwtFarabxeKx8Pm+0JKzynjIhYbvd1rNnz15g8NNGuVzWcDjU1tbWC16H9NxTg18ajYaFktKxAU0mkwYhgaWhHPr9vuHTg8FAh4eHxofhZyQSCbVarYAH6DF6YKh6vW4QmY/GEDqMCKF9p9PR/v7+RHSZm5szfiEqYq4ok5mZmRc8eQwAuLyHALwxQVn7qIhwWJKt20M0XqHDc94jxNP3ob3Hxv2YmZlRt9tVs9k8NQ8UHvPz8xoMBtrb2zO6eywciKDX61ki1PMLMGUkEjF+JjrEcLNvRDCSArpjOBwaDEjUihx5hxKos9vtmn6BlnyG/YJO+Xw+EPGfNE5VwPV6PZAN9UznQxgfMvhkmA+L8EDCrrwksyr+2dKxJ1csFpVOp1UsFg2PjsfjqtVqWl9fN6EgVJWOrbKkgGBJxwqSUKpQKATCv1cNcDGEH8+GQejH70+ii/fYvbfmk0JsbtgLwSDNzc0plUqpUCgol8sZw9br9QBdvHdFSOfpcVJJYSKRUKlUOjNNpCC/nBTmIxSEad6z8cqEAe08/gZDw0PQFrpMTU2pVCopl8upWCxqdnbW6LK/v6+VlRUdHR2pWq0G+AUhOgkKwQtCgXkv6SwDGAbl4MvGJAWEFDgNxeuTUJICeKjnF57jE03QFtilWCwqlUqpVCopn8+bYTo4ONDa2poODw+1u7sb2AP2DPkNJ4W9MwTPn3U0m017LhVAPipEd6BfmI9Pkvm5ev6AdsAYfoT5Bb1SKpVUKBSMLp5fDg4OAp6+5z9oz99eztFVL4NmTlXABwcHkmR4LvABg0RA2FtjEnzGe7Xdbtc8GW9tfWjsrTOKoFgs6r333tPNmzetlvenn37Sb3/7W+3s7GhnZ8cYAMPgmdh7NslkUrFYzJh+fn5+Istdq9WMuHic3ivwJVMIHPPxuLcPGVGMYTxPOjZ2KCKseqlU0tzcnN577z3duHHjBbpsb29re3vbEhej0ciUgc/Me7wUJTEzM6N8Pn9mmkAX+IXyoTAv4KXg1SIseBDwD7TF8wrzGMMnzBKJhFKplCqViubn5/XLX/5Sd+7cMV579OiR/tf/+l/a2NjQ9va2RXXSMYRDrS5KXpIJZDwet3dMMsBHEUbvJUnHGLXnF4wmcoEi8jkDj4XyHGiCHGBMU6mULly4oLm5OX3wwQe6deuWksmkUqmUfvjhB/2P//E/tLm5qe3t7YAz4g2DVyy8B4OF8ptUjiKR52WnJNdRvsByREFEciSzkSOgI2hL4hcZgU7hqJqoIplMqlKpaGFhQb/85S/11ltv2V4/evRI//N//k9tbW1pc3MzoNg9XcKQIjAE856fn39pkv9MZWgs1ltt73r7Uguv6T3W5z0/b8EgJsyH0o3H41peXlY2m9WFCxeUz+e1vLys2dlZY7rZ2VktLS1Jep6t9EA8TOFxYc80MLUPa846wnTxIQ0MwUb5DZGOhcR7ft7rgi6e2aE7ofXy8rIymYwuXryofD6vxcVFZTIZUxS5XE6VSsWSoV4owl6KnxsMBANTD3nWAV35HsYX4+GxOW+IPD08TolQQSfP4HikQBTxeFzXr19XNps1uiwvLyufzxvvzs7O6uLFiwZJwSeePqwXHvN4PgZwkooZ/1yvWFi7X7f/XNgZOYlfkCufnINuCH8ikdCVK1eUy+WMHvwNfaELCWCex3t9riAcnnvFM2kVBM6GlyP4hagYmiFLYTkK752XTem4hBaFOzU1ZTmJixcvKpvN6vLlyyoWi6ZvoG8+n9fFixclySqewlEr8/E6wBtN4NmXjVcqYLwV3G4pWEvrGYIsMxbeZ1mTyaSFQr5iAgs4PT2tTCajdDqta9euqVAo6L333tP8/LyFTr7iQZIWFxctyfP9999ramrKFIz3usCC/OkalAKCOQnjgDuhaPBYKJfzyov5eijCewqsHcOB9wzdPV1ef/11zc7O6s0331SxWNTCwoKy2ewLiZFSqaRbt24pHo/r/v37AaXhKzPAmpkrSvPo6EixWMy8wbMOwsTwQYzBYGAlU/CMX7P/46EdnuN5xnsWQFPXrl1TPp/Xhx9+qIWFBeXzeaVSKeM5PKlKpaJ33nlHyWRSv/vd7wI0w0CgWHxJEqFxq9UyQzjJgLZh79qXhYUNsHScMPL84kunTnrPzMyMstmsMpmMrl+/rnw+r1/84hcqlUqan59XOp02Bwd6VyoVvfvuu0qn0/rss88sCoJffALeRy94ghzKqlQqAUX0qkHpIYNQneobIhfvlDBvH0HyO4yAT75ioNm3TCaj27dvq1gs6v3331epVDI5Chu/xcVFvfPOO0qn0/ryyy/VbDYDBQkeDcCYebiQyg4ioJPGqdQiIQDTh0NVj7+haLyC8YA6m+0TEHwvmUyaJUqn07pw4YIymYzy+bzS6bQJqsdZUN6ZTEaZTMYEBg/Uew4oWw+FhCsTJimfIWHHhnnFgUJmeHqwBr8WlKDH06XjxKKny9LSktEFjzeMnTMPvB8SJDzb478+Y+yxZ6w7+3/WQR0ztD8J5yb0hi5+zSdVG3j4BuwYGrz++uvKZDKqVCrKZrPGL95D80aNBCMeEAlQDKjnG+bvcxfQighnEn7xaw57bbzX50j8PJhLeI+9AsKzy2azunbtmjKZjJaWlowu8EuY9zAuGCySpD7H4z11TyuvC+CdSZKTvnpF0gvrDitazy+et3iGl2sPKUGXS5cuWeSYy+Ve4BcMpYf/4JeZmRkrmQvTJIx7+3mE5xkepyrgarWqSCSiYrFo3oDHaahnJEzEYnuiYsG9dfNHSpPJpMrlsq5evar/+B//ox3GiMVillhqtVpqtVrmCVJfGovFVC6X1Wg0lMlkrKSMekJfbcCmJJNJewfeB2HHWcfu7q7ND9zJ49a9Xs8Ku6GLrxOWjrFxf+wahQcMs7CwYHShMD0SidghiV6vp3q9bp5eu93W4eGhpOfHPGdnZ5XL5TQYDLS9vR2oXfaM4pMF4ORk7ScZ+/v7FmXE4/EXMvbsC/gZlS/ScT0sCoHKEI7rEkoCwbz22mv6L//lv1g/gkgkYhUeh4eHOjw8tMMFR0dHduCjUChobm5OxWJRo9FI9XrdPC6vPFBs4KQoPcrrJjFMVHyw//7gDkX8viqFbLvnF2+wyL73ej2LKpLJpJaXl3X16lX90z/9kx0Lj0ajpnxZJwqDI+KSzKiVSiVFo1FtbGyYnA2HQ4tYMFo+/0FOhVzDWcfOzo6i0eODXr1eLwDz0E+B6hk8T+lYv/hkt88xeTlaWlrS9evX9Z/+039SOp02GiK/1DpjhJAjdF+xWFQ+n9dgMND+/r7x5EkQq8ezw6WNJ41XQhAeK8UyADt4jDKcGUahecggjHPhzQCCz87OGqEJ8SHkSZgUG5VMJi2Rtr29/VIrzBxZh8/uTqKAoYuHFDAafr7eK/ChEn/zM+9JwNieLrlcTslkMnBg4qSMtFesMN/8/LwkaXt7O+C5+fczvOcQriU+ywhDMuG98jilTyr5gQL3dGGfOYW2tLSkSqWi2dlZpVKpgLLytGdNPIPnYNzG47E1HIIm3rPDO/Veua+GOOs4SY4wfD7s5XdejqAPUaAf0BO6VCoVVSoV8+x4H9Adyspn78N0ofJlY2MjkNRj8P1wtBCm+1kGvIBX75WZj8p8UpG5eLp4jNrnqnCuFhcXValUVCgUzMOXZPAGcioF9YuXo1KpZPzi6eLnw1p8BIeee9k4VcKWl5ethtRbTvDB6elpOybLZoCPYpHAZAh/ySTjid65c8c8XzA/akFZXC6XM7wJK0uB+ng81tzcnP7Df/gPWl1d1U8//aRarWb1iydhjO1222p0/QaedSwtLWk8HqvRaFjtMokdsv8YEiy6pwubz/ypVSVcymQyunv3rv7u7/7Owh/gFi+w3jvy9a2MxcVF/e3f/q3W1tb07NkztVqtF7pa+QjBwzeZTEblcnkiJXzhwgWNRiPVajU7MSkd45bMz1c/sDaE0dd54vlCk2w2q/fee0//9E//pGw2q5mZGUmy4+coutnZWUnHBw18HedgMNDCwoL+8R//UU+fPtXDhw91dHQUiGbYjzBuDgRSLpcDDV5eNSqVSsDbRsipNvHGDuHGmPl8A8qPnAY10alUSnfu3NE//uM/KpvNWo0qDgzv4BnII4YepbG0tKR/+Id/0LNnz/Tdd99ZvbzH6j2/+FrvaDRqtfdnHeVyWaPRSK1WyzxKSWbkGB5C8o17vCx4GQB+S6VSeuutt/Sf//N/NrrwbK+jeFe327Vj+UTJ4/FYlUpFf//3f6+VlRV9//331szLVz94aJPIj8M2r7/++kvpciq12MCwVfNW0yfn+J3PPEJQrA5ESyQSmp2dVbFYVKlUsgylJyoEIFzH9adHBM8jY9loNF6oIvAJOekYQ/P4DT876yAJ4q1f2CNFyXm6eIvO8BuIApqdnVWhUFCpVDLFFc4GS8eJI0I36M776cjUarXs53wm7JmehFVNGhmQLIOuYe/Wewn+Hf5v+MbXLqOEi8Wi5ubmtLCwYDWcYQ/S0wXFiQFgTvF4XMViUfV63Z7hPTieFcY1fT7k59DFe56eHuFIxv/O0yeMHSNHhULB5CiZTJryDtMFpYAx8ZElDkCxWAzIkecPPycvMx7XnsSZwUE6iU9Ys3R8JN3/zMsT2Ln/fZgur+KXwWBgBzEoo+TZ8Euj0TBF6vfOzyUsQyj8l3nBpypgyiewsr5Vnw/lY7GYhTw+uy4dA+n+/91uVxcuXNCHH36oSqUSUKJkm1nIcDjU7373O62trdliK5WKlYwsLCxYRrnf75un5GsIw0zMfCjFoVnKWQenckha+OYqMCP/p2YUj5M1eMwYZd3r9bS4uKgPP/zQWkLG43HlcjmjC/QejUa6f/++0UV6Xv1QLpeVy+VULpcDSpZkJZ6Chz98SIxQT09PT9xesF6vB/jFNyjxvAAu6SMnFC+Qjm/YMxgMdPnyZf3617/W4uKiMTR0oc8Cz/vkk0/07NkzW8/i4qIlXhYWFixXMBwOlc/ndXh4GKCLP+oNb/pyy/39/ReUxmkDnBW6hNtDSsHetR62CivmcJ+Qy5cv61e/+pWWlpasRUChULAIDXkbDof64osvtLq6au+ELpRzTk1NGV6dz+cttwH0Ar/gTHncMxKJaG9vbyJ+oToAOfI4OEoX/kG/+N97OcLhAF+/fPmy/uIv/kJLS0sWLWSzWY3HYzuxhwGAXxjol3w+b2WuyBK5FXJZKHX/GfYJpb+/v//zPGDA7HBZkcfKeKmv6/RensfUwKEkWX0vJWCETHwevKrX62lzc1OPHz+2cpJYLKbZ2VkLJaRg8xK6i4ULsBkwFM8iHD7r8PXDrBnl6mmCtwRWHB54ZN5bptrhJLr4DPlw+PzI75MnTyxSQdH65tHQE7p44YcuYW/YN0CaZLAGQmDv3UvHOLWvseU7nok9v7DuXC5n9b0oKp/V51ndblcbGxt6/PixVcbEYs+PhRKKo8CAFDi2HE7aMhB2eGmSY+usG8Ue9uDCHiP8EvZ2GZ6nxuOx1fDiTPhkqsdXkaOnT58ab0xNTRkuCl2gJ0qL75+kZNgHHIKfwy+sJxqN2sEYz5c+SewVLiMsb8yFenA6HfIMeJCout/va2NjQz/++GPgPblcLiBHHjYCqsIZCusOHykBnb7spO2pChiQGsblIXhrvsbOewk+bIO4KAkE4fLly7p06ZJmZmYMt6QpDtnXhw8fqlqt6vPPP9fKyopmZ2eVTqcVj8dVKBRMMXU6HW1sbGhra+uFBBJKy3sQuVxOnU5HtVpNiURi4r63CLVvVA9dyBDzO5SZx9H8xgExsOGXL1/W4uKiWc/RaBTwZI6OjvT9999rf39fn332mdbW1pTJZAzf4ugtTfPX1tYsAefrb6ELHjleZb/fV7PZNDpNYphgWLBAj1l6fvHn88PDY+SeLleuXNHFixfNW4pEIqrX67b/nU5HX3/9tXZ2dvTJJ5/oyZMnmp2dtQoA6ILHvL6+rq2tLQvFfR0uissraBr5cwR8Err4UDZcH+2VEPvjk0oekvBJ3kwmo2QyqYsXL2p5eTmASR4cHGg0Glk2/8svv9Tu7q4+++wzra6uWhMgSZarWFhYUKPR0NOnT7WxsfGCwvDYMfmOVCplx7o5UfZz6oBxpLx3TwN/nw+i1M7LkYdJ4f9kMqmrV6/q8uXLdngpEom8oF+++uoro8vKyoqVJ6KnqKhpNptaW1uz03A4NH6P4Btk39d5nzbOdBDDY5AoYhjD/w5mgzhM1DNOOp027LdQKAQ8tcPDQ7NKnU5HKysrWl9f1+rqqjY3N41haaiDl9zvP+/122g0TOgxBL5cRJJ12aeEzq/rrMMrVW/tgD38ev1zo9Fo4Aoe/5l0Oq18Pq9CoaDZ2dmAR0IpEOUyKysr2tjY0MrKijY3NzU3N6dcLqdWq2U4OTTc39+3Hg3MEw8a74K5+aY3zHESgfJJDQ+teAOIosG4nFRY77G9TCZjmHihUDB6kUzlu4eHh3ry5IlWV1f15MkTbWxsmOfBrRYkkgeDgfGLz5yHcXyfxUZoPX+ddYSxY56PEfTVPmEI4qQ8Akki+IWogGfiSWKwnzx5opWVFT1+/FhbW1uan5/X7OysDg4OVK/XNTs7awmoarVq8IwfXvlCFwzT0dGRKcfTMv7hQRIVY+ZLy4AJkTXvtWNc+I7Hq8kVUDrm20nitXLY6KefftLKyoqePHmira0tU97lcjnQKhPjS+MmImh4yTubJDs9Rn5ahP1KDBj3mzpC3+wYpQmRfBiC+024hkczPz+v69evW0exZrOp/f19VatVPXjwwDa03+9ra2tLh4eHplgpzbpw4YIuXbqkdDqtWq2mtbU1ff755zo4OFC1WrUKB1+z7MNdlA/NrulYdtbRaDSMLtFo1OYryao+jo6OLBmC1YWRotFooG8FXhV0gcFqtZr29/f17bffWi2rpwseIHRZWlrShQsXDK/f2dnR/fv3rb+v73nrFaQPV8N13q+y4GG6+KTO4eHhic33eb+vA/bHl6ELFQdvvPGGFhcXDdfc39/X7u6u7t+/b7Ws/X5f6+vrRrfRaKRcLqelpSVdunRJ169fVy6XM2/mt7/9rarVqg4ODixaGAwGxqcoAS+0RCX0IzjraLVaAbqwj+QRUJZEKcPh8d1n4TpgYItKpaI33nhDFy5csGfu7e1pb29PX331lfFLr9fT2tqaWq2WYZ8c0rhy5Ypu3LhhtfVbW1v65JNPVK1WVa/XTY6oXiKKo2690WgE6ML7zjpoyA40A2yIfoF2RE5h50A6PsyB91ypVHTz5k1VKhXjyYODA+3u7uqLL74I9PJeWVlRs9m0nr2UrF26dE5J1EEAACAASURBVElXr15VOp3W0dGRNjc39fHHH2tvb0+1Wi3QI9nj1chSuMzQOzrhcaoCJtkU9vCwdL6AHWH2Nz/EYrEX7tjK5/O6cuWKCoWCKfC1tTU9ffpU//2//3c1Gg3zhMESEQKuySmVSlba02q1tLOzo4cPH5pixGv0AoW3IR2faWeTUQJnHdBFUsAwccQTq+np4m9+QNF4bG12dlaXL1+2xNLh4aE2Njb07Nkz/bf/9t8CJW8eM5NkHa7m5+etjrPb7erg4EDfffedeXow7HA4tHI06Rg/DyfEUBJnHRgyPFzokkwmjV8wyF7BeagmTJdisajXX3/d6pkPDw+1urqqH3/8Uf/6r/+qer1uhsgfCIB/4JXl5WXb652dHX355ZfWVAl+8eulYsDTxTdPn2SwRowenjkHaPxFBwhyuOmMz79EIhGjy8LCgnUG3Nzc1I8//qh/+Zd/Ub1et3adhNXwTyaTUalUMmVDpFitVvXNN99YkxwMAbkgcH28bA/rsd+TRAYnOXicXkOOOMjl5QhPl732JbDz8/N67bXXND8/r0gkYvzy+PFj/cu//ItqtZrRwcNN0vP8C/yytLRksN/e3p7xi3TsaDJXngG/wL8+ejwJbpNeoYBhejR6+CFYHZ9M8PWREBZcBOwvl8tZ+NHpdLS7u2uWBbed5irgkJFIRG+88YYuXryohYUFRSIRNZtNra6uBoqjfalRuDCf/yMIHm+bZMzNzQWSP+GkG//25V4+YQid/B+OS2J0er2e9vb2dHBwELgGZmZmxo6ash7oMjc3p2g0qkajoa2tLa2vrwfgovF4bN24fLkeYRod76id5TLJsw6K1fFUTkowsT6fBPPRBPNjTqlUSrOzs1Zf3W63tbm5qb29PaMLeYobN27Y6clYLKZbt27pypUrKpfLisViOjg40MrKin766adAlzX2B171CZRwiOvXcNZRLBYlHVeBhLFdPDxfVxpuhASvejkC36bhy8bGhnZ2duyWGmhODxGcqZs3b+ry5cvmJdbrdXOC8GKZDxUx0AVaUKXiy/0mPSEIXTxExfC090oSvSHJFDP6BZ5Fv2BAtra2rP2oT4i+/vrr9tmpqSm98cYbWl5etsosciirq6uB6i/p+PYYD7uxZ8wZXfOz21EuLi5qOBxqd3c3cBDDWz6aY/tQ008U64vSy2azhrVIMgIhUFimRCKh27dvW7lZIpHQhQsXrOwM3Pfbb78NgOMwre9wNh6PjaFQ5mA1GJdJxsLCgobD4wbbUvC0FkYIpvTzwYsh7IJxOLOP0ut2u9ra2jLa+9DszTff1NLSkn1+eXlZ8/Pz1hCo0Wjou+++s1aUfn0+cchcmAf3yuGxTVpYT3MklECYruzBYDB44UoeFLevv4xGo9YHmlC92Wwa9u2bhFN0f+nSJRUKBaVSKV26dMnm1O/3tb+/r88//1wbGxtWYgVPhDvi+cMjGFKST5PeIQjPHhwcBA5i+ARouCoEHvKRFvs0NTVl2DiJ03q9brkBDjawLujCicqLFy+qUqkYv1SrVf3+97/X+vq6QTHsVSqVOrGmm+PB5Bs8zDYJXYbDofb29gL6BS8/zC/IjXTsFMLP4VyKhzBWV1dNFvBOk8mk3n33XV2+fNnk6OLFiyqXy+Z512o1ffPNN6ZfvOEMY93Isa9gwTBwI8ZJ41TparfbL5QE4T0hxB7b9OVWvk4PLwsr4S2CnzAZy/n5eWtCUy6XrRMaPUH39/e1vb2t1dVVbWxsWLNkLFJ4eMAc78YnOXwpy1mGT8aEcVS/JoyJXyOC67PJ0AXPjWdAa+jCseRLly6pXC5bEyI60B0cHJjn6+niqy78vPC4CJt8LewkCoYBdER1B//2ySUMpd9/6TjBxHx8MyF/ZJTvYOwTiYTK5bLy+byuX7+uxcVFq5Shjnhvb0+bm5t68uSJ1tfXtbu7+9KDN8zXR1Rh3giXib1qILwYaOYf3gPpuAzNH3f1BhPojNOkvgSR59LTACN99epVLS4uviBH1WpV29vbevr0qdbW1qyO1ytbz4e+4gd54p0++jvr8I3qgRz8moE78HR9Yos1e0Xs/3g5CifgaXtw7do1VSoVkyPyFPv7+9rY2NDTp0+1srIS8J79c0/an5OKEfzBr/A4lWL7+/uBENVngAHNacbj74iDkfDkxuOxlb7AHL4sCwLjGX/44YcqlUp68803LcNLQq/X6+np06f6/PPPtb29rT/84Q8vVDT4JBdKDdwNhpVkVteH6WcZHDhAyL1yZc1EBL7xeJgufAcFSzIjXAuLR/fLX/5SpVJJt2/ftrpPntXtdrWysqLf/e53Rhd/tNIbRN6L8YC5YBJf2zgJXWjeAzNLCvQWJmHmmR2sk+PECOXs7Kyy2awdLgASYE6UHGUyGf3VX/2VyuWy3nrrLRUKBZsPHt7jx4/18ccfa319XQ8ePAg0fUGAMKL+KiuvEBjekJ91kDMA5oHuPj+B8HpP2IfheM7xeFzpdNq6eXmITXrumebzeWWzWf3N3/yNFhYWXqCLr47gkNM333xj0ZKvAoEnfBtMjPrR0VEg6oUfJ+EXvFGgJ+lYjsB8KVX1GDD9v6ELa06n04YPo3ypQsjlcspkMvr1r3+thYUFvf3220YXr/CfPn2qjz76SBsbG3r48GFAvwCZodfQJ+TH/H12OHk0DjtpnKqAUXwwIP/2ZUPh8h3+zYAI/ve+dM0fPKAfBAc0cPNhUDDRjY0NVatVy8J6LIZ3o+D9//3fDELAnzPCa/LZbOhHZOAZ03sRPAeFyOeSyaQWFxfNcEEXehZg/QeDgXZ3d40u+/v7ajabgaw67/Dz8Hv6MsGZFJpheD7wCYqwVxf+N+/05Wh4ifw/k8no0qVLKhaLhpkvLy8HDhQQakKXlZUV7ezsWBkRdPF86pWY51c/EKpJFbD3bv07MYY+webzKYywBwVd/bOAooAZaOtaKBQsUcQe7OzsaH9/X2tra0YX31XspOETo2E6+blPkrT1NPTG0GPl3hkI79FJzwnzTCaT0ZUrV9RoNMyYwy/Ao+QsdnZ2jF/29vZUr9cDN27496DgvcfrPxOe38vo8speEOGFSccN2VEaWHaPkfhwxStowHsIvry8HLh7zOM8YHfNZlOdTkefffaZvv32Wwud8PzAfryHRxjiQ1o8D3/dy9TUlBV7n3V4heBx3E6no1arZWFQOGnhPVsvZDC+vwm2XC5bwT+C6z2Dfr9vJUKffvqpZa8pw6NhPkehfaN6mMQnSaVjgYIWkyZVvDEiOURBervdDvCLP2XFu8PP4rMeF75y5Urg+Lr/m+fUajUdHh7q448/1tdff63t7W3D8fB8qUYJR2zwMPgp8/L4LJ7fWYcv3fQKjkodXzfOc3Eg/OEn5uENN+teXl42fvF1xRjqXq+n/f19tdtt3bt3T99884329vbsAlWMCrLny8nw8FhHuL4bPjrptu7TBryJc+CrcuBfv/eSAhc+hGkAVOL1y5UrV7S8vByoTiDPgtfaarXU6/X0m9/8Rp9//nlAjjhAlM1mJR33Ag9fUuH703hePs3BkV6hgOk8BmE8thk+GcP/w5gZk/EF0M1m0yYMwbzV4hkIJ9+hhs+Xv4Xf6f/m937uJyXd/AafZaD0w6FaGE/13ozHrsJ0IbRqNpuGkaOsEFyPVzM6nY71voUmCLWnhRQMo30GnhHGqX0YfNbBIRSUq1+j30/C7DDswTygJeVY9DymeTZVCN5Qez6kRrXVahnPULeJd+1Dbb/+ML94moX39qyDMi6/L36+YTqdJGMMIKN2u210AaLhbxRBuOLi6OjIyjyPjo7ssEG4Fjw8wj/zUSTfY98moQtK3ss9z/T0Zl6ehxjsJwbhJLqEI1zeBW93Oh01m0074AXP+PWEo5Iwb0hBp9PrpdMiyVMV8ObmpmKxmJVqgHtgXfBwcOV9/aL/2Xj8vIC+1+vp22+/1Xg81vXr13Xz5k0jqmdGEjUoIk7D1Wo1s4DUTbJoGihjbTjbj9dBdjccHodD8rOM3d1dRSIRO1btaxHBn+naBg1oPOL7akiyq9O///57jcdjXb16VTdu3AhgjSQ6Sf7x3LW1NSuy51AKDZTwfPx7wcxQPsyDJB1GhXdPWh+9u7uraDRqDfxhYmjd7/cDkNFgMLBaUM8vo9HIwuJvvvlG4/FYN27c0N27d21u3uP1dBmPx1pdXbX2m+l02rwZ6fi0FXzqa3vx5OBtvB++42tRJxm+2ZM/5IFzw6lFXzdOAx+iM+QIPv7uu++sBPHu3bsW4WAgxuOxJUXhF05PNptNZbNZ7e/vm9wQinvsWTpOIHrHAWjAR8QeXjrr2NzcVDQaVTabtYjJOwcceJiZmbFG6tDF84ukE+ly584djUbHl/4CG5D7IT+0srKilZUVtdttzc3N2RFunKDR6PjkHfxyUk2+x/GlY2frtIM7pypg7xV4L4bEABvtrbW3GkwIb6rf76vRaGhvb09zc3MvnZjHAdkMQofhcGhn8r3FDH8/7B17bwviTJrN9nTxVt9jeGxyOELw7/HWlBC02WwaXWAY/1k/V08XkiC5XE77+/tGFx/as1avtBC48Fx9OBe2+mehi1eQXljDz/V08TzAXgNd1Ot17ezsaGFhwYTA08bzZtjjTyaTyufzqlar5pWHh4cY/DO8BxaOkDwtJ6ELz/JJN5yBcAR5El34+WAwUKPR0O7uboAuYW/0JDmSZP0UqIrwtDhpz5mTnwNzD79jkuE9f/jFV+zw7nCEGX4XewVddnZ2VCqV7GAQz/aJOQ/jMFKplAqFgg4ODgL1xuxZOIo8KYLx3nU4yjtpnKqAuSmVE00oXF+gzcv9FTseH2WTsFabm5t2UV2tVrOjg0ARnkAQ7/3339c777xjZ/q//fZb5fN57e/vW3s9rDGejseXpONCe57pDwtMmjzg9BB0Yc4ky/zBFd99n81EaTLv4fB5Z7N2u21QRLlc1o0bNyyk9MJJwfe7776ru3fv2im5hw8fKpfLqVaraX19PaCI8RR8eOoz8WGG/DnYOK37CG39oQLK7fDGoAv/94lZRrfb1erqqvHL/v6+lpaWdOfOnUBljq9EiUQi+uCDDwL88uWXXyqfz2t3d1dPnjyx/WB/4FvpGAagZIm99dUG4UM1Z6EL7wJXpOaZqgv2FBnjHfzch+iDwUBra2vGKwcHB1pYWNDrr78eaHDvS96i0ajeeecd3bp1y0Ltr776yjxhDjOF+cWH7+PxMU4bTgTG489vV54Egrhw4YLJEV70aDSyPJGnv+dfH+l6Z6/T6Wh1dVXtdttKMpeXl/XWW28FShl9xUU0GtV7772nt99+2+CZL7/8UtlsVtVqVU+ePLFIyRt7f96B/fJQrVfCp9VHn6qAsZS+JwAWCCJ4L5CN8QotLFSEcI1GQ9Vq1Toi+VpRnwyJxWKWyc1mszo8PNTBwYHm5+fNe4QxvPUOW+iX4byTenmeLuG2hDA9StWHgx72YH78n2gCrPvg4MBA/zBdPI1pNpLNZq0rFb0kdnZ2AhbZQyQYQzz2cOkb7whn7c9KF2AOnomC9R4aChOm9YlN6VgBkDfAs+GeN68AvMckyfhldnZWh4eH2tvb09OnTwN3lkEHwlHeDT/7ulcPo0yaL5BkkZu/zBJa+/p6KdiC0a9POo5cPF3q9bp2d3ctYegNadiDLhQKikQi1nh8d3fXerJsbW3ZZ8ORYjg56X/n8U9fx36WgWPh5Qh+4TneWfLzgT7hulyS07VaTTs7O9YD2NPEz3s0GqlYLJoBOTo60u7urkqlkuU0vE5i73yCk2eH+YXnnzZOVcDgQ9So+qYz4RDSK+OTwv3FxUXNzs7q9u3bunr1qjKZjHK5nN167IUUgvMMKi44sst8vNcQxmj8jasoOI6e+uOlvg7zrIN3ebr4sNt7EtAl3OMAxQpdbty4oStXrtiJOH/c1QsAGeLxeGyVBZxKwgP3J6u8Bx6JRAK1nsyNagdfDcGcJ1HAYbp4DNhnzD2/oGjDkAklVO+++65u3rz5Ar8gfJ62GDsULTWz1J6TDyDz7eniMVgMKBiw78iFgZnEcKNgwg38qU/33rtPgErBY7rRaFQXL15UoVDQ3bt39cYbbxhdUqmU0ul0oI8v9PY15/BLOp1WOp1WNptVo9GwaJbP+jpu6ABdcHjYQ/592oGDV/FLKpWyY9Deq/VYOfzinUD4BbrcuXNHr732WoBffBQzGo1e4Bduk5mZmbGLCyhZm55+fokq2DM6x/et8EllDFFY/71snOlaehrreLAdYmMN2ACPofmwidsIXnvtNd25cycQdnmPGgKDAcIUeHwUbfvz6cPhMFAo7ZM53nvwlQk+jDkJVzpt+E5WHgvnPeHCeoSJ3/tNyWaz1kCE0Nrjld6TgeZ4mAgUR7V9bwnvSWJE+X84ceq9Du+dToqPQxeMhzcAXiGEhdnzC+/M5/NaXFzUrVu39P777wcEns/6MkMSWYS00vNeA5yK8xCUdFzh4/voIohAJChJH1qyzkkUMM8hfPa08pGj9KKw+v2IRp/f6ry0tKTbt2/rvffeCyTDRqPjpJDH4/0Rf5LqHIbh4IDnMQ+znHSzL/IT9swnrQ7x+sWXLYYTftJzQ+kdJeYZ5pebN2/q7bffDlxDFaYxew1dOKgxNzdn8scJTObCZ5AtlDgQmlfARI4+6f+y8cp+wNJxaO3Lfmg8jiJEqHwSgdNK6XRa77//vjW74Oohr7wlWc/Wra0tNZtNff3119avlLP+N27cULFY1Ntvv61YLKaHDx8Gmod7ZgqHddJxqIQyoKxnEgXsMTbKrsJ0gR6+ttHj4YVCQZlMRu+++66uXr1qZ/WZq1c4dBGDLg8ePDC6JJNJ3b17Vzdu3NDc3JzeffddPXjwQF999ZWGw6HhuAgLVwFBD35+EtNPOvAUfTN5BNPXmVLf6ZNnfBal+e/+3b/T7du37ay+D9sx7JTera2tqV6v6+uvv1atVjN++eCDD3T79m3Nz8/r/fffVywW01dffaWpqSml0+kAFs91TQwUNuWS0MXz+VkH4TJ76j0v7+H5+miiEOCQQqGgbDarDz74QDdv3tSlS5cCF7D6vQSDX11dVaPRsE5e3Jb83nvv6ebNm5qbm9M777wjSfr88881Gh33VoAffDQVPj3IDSu+C9okcuTzRZIs34Ecgb2GDyl573J+fl6ZTEZ/8id/ojfeeENX/k+nRc8rvvqh0+lofX1d9XpdDx48UK1Ws5Om7777rumX999/X9FoVF988YWi0ajpF+Q5m80G8l/QH4eQff+/SsKxEScpGsJ5f2urD7+ZHCH1rVu39N577wX6rfpbEsBuDg8PVa1WVa1W9emnn2pjY8NOypXLZV2/ft0EbG9vzzYH4+AVCkT3gDufR/GiICYZ3jChiCXZkUW8Cl/L62sy2dBisagbN27ozp07gVaW1BnTWZ/kXLVa1d7enu7du6fNzU27mr1UKunatWvK5/PK5XLa2dkx4abo3CfXfOLD45reYP0cbBxogAjF37KMx4LBZm38jj3I5XKam5vT22+/rV/+8pemmOAXj2HDLzs7O9re3ta//du/aX193U4MXrhwwQQqlUppd3c3oOh88g8vOHzXF54Q6/g5GX9Pe/4Q2dF5jDkQvvpoUDqOlG7duqUPPvgg0GjcQ1usgTatOzs7+t//+39rbW1NS0tLdknltWvXlMvldOPGDbvsgPd6jJVEGM2LwjSDzyatGZcUgHa8ckM2/W073vDhmRMVz83NmX4BoqTkEaUOjHB4eKjd3V3t7u7qo48+0vr6usEXpVLJWsLevHnTbkzBUHvM1zuN6MIwHs0I1+/78cpmPN46w5zSMUZEN3xJZiWl4wsI3377bS0uLqpUKikafd4qkQMEjUbDcDraDDabTf3hD39Qo9GwZwMvUB9Mxy4f3npsUVKgtaUHxCmH81AKdaRnHdwyjIL1988R2uMJw6TMlQMFd+/e1dLSkvVzPTw8tGPE9Xrd6EK7zmazqW+//TZQ6+sbZvP/drttCpz98MYTvBoBgtlZD57Xz0nCcUKMd9BPwYfJ/mQXng78gidDF7OpqSk1m00dHh6q1WqpXq9bJEakVKvV9NVXX6lWq1njc/oFcPgCOvlkDwqDNXsc3+cFfHLFHxyYZEAXDBNyFIvFApe24j15OQKT/JM/+RMtLy8bXTh8w5U5RBc04vcRAfza7/et7tzLEYrKK3xJZqg97svc4J9wF8JJIAj2ySvVcPWF79PiMWr45YMPPtDFixftuqp2u21y5PUL11DV63U7NUpPF/SchyQwjsgIHjn6BRhQOnZqwrwOTcIJUT9e2ZDdJ8hQNL4he7vdDuCueLXpdFpzc3P6sz/7M/Nao9GoZSc5BkmtZqPR0A8//KB6va7vv/9enU7HQHCytiyaJApEgSAwNMkmvFHvucdiMdtEQrxJkyrhGw44iJFIJJROp+3deE8oI0q7isWifvGLX+i1114zPA4vbnd3V1tbW3Y9favV0pMnT7S/v6+vvvpK7XbbcEu8EjzNo6Mj1Wo1M1y+RhkacOwym82at8XcYC7WNWmPDF+8PjU1ZUlbvBJOcAE9gcPFYs/vOCuXy/r1r3+tW7du2WGOg4MDra+va2dnR5ubm0omk5qdnTV+2dvb0/3793V4eGhYIv1wvSHgODsDDwovDEOWSqWsixrQlBdCFOkk42WODL1PaI5DFOKNIrmTv/zLv9TNmzdt36gKgV9IIDUaDT169Ej7+/tGFyAh5AdFw4lBD5uh8FAu0JKaYdpPSsFEPHpiEoPNIRyUO/oFQ0VlkNcvnl9KpZL+4i/+Qrdu3bKe0Ts7O9YXeWNjQ6lUyvTLo0ePVK1W7WYM6MLe0wiKssdOp2PzY39oeQBv+0RdGGv2pYsvG6cq4Hw+LylYbsILPCOGkznxeFwLCwsqlUrWMIXs9M7Ojn788UdVq9WAoqGkptfrqVwum6cqydrqlUolxWIxuy2Cywel40YneCnMBYb2iQUGIXO4F+yrRi6Xs3/7RJIPpU5KciUSCVUqFZVKJctcS8+Vwe7urp4+fWqtE2m4TV/Sfr+vcrkc8KavXLmiubk5lctlM4hhurBGD7P4JA208yEU35vUA4ZfpODpKF+OBq19NUYikdDi4qIqlYpVgIxGz2872djY0A8//KDd3V2jSy6X09HRkfWRrVQqgWbbNOtZXFzU9PS02u221tbWVK1WbT98uOvL0NhTHwGE93jSngeeX8JepnTcEIrkjp9juVxWpVKxy0mJ2Lj9ArrQMazT6Vif6oWFBfN2I5GINe2vVCpGl9XVVVWrVVub9/h93oTf+dwH9ApfjHnWkU6nbf3haBWDxfN8Up0WpEBw8At0+cMf/mD9PyhH7HQ6qlar6nQ6WlpaMuxakq5du6b5+XktLS1pZmZGu7u71obSV3nBG55ffIniSfXhwLMvG688iDEYDKwpOALDywmnsRiSzPvlHq/FxUXl83nr5fDo0SN9+umn5gFjufFGUqmU3nzzTcViMf3000/q9Xr61a9+pZs3b9qljNVqVV9//bW2trYCBdrD4fFFjf7UEdiWb5ZBqADTTCJQS0tLGg6fd9vyWWKYxl+v44/JZrNZ3bhxQ0tLS6pUKlbX3Ol09PjxY3322Wd235kXdITrzTff1NTUlNW0/upXv7IkUyKRUL1e18OHD+0WZC8QeDbMkVAS5YswQQdC/UlCyuXlZasp5ag4wjUYPL98MpPJmKcuybyZN99800LsfD6vg4MDtdttPXjwQP/2b/9mHnA8HreqHPiFqpqVlRX1+3399V//tW7fvq2FhQUlEgkdHBzoyy+/1Nramu0XuK6vKPFK2ScK4TGiAn+d01lGpVLRcDi0I9G+8kM6buRChMDPU6mUbty4YQ2rMpmMNRp68OCBPv74Y2tM5T1syqju3r2r6elpuy383//7f2/XGCUSCVWrVd2/f1/b29umVKCLb14eTijzt88feON61kHNuu9qiJLHYcKxaDQa9m4w2gsXLqhcLiuXy+ng4ECHh4f69ttv9dFHH2lra8v4hfaUwBFvvfWWpqen9ejRI3U6Hf3617/WW2+9ZQm9g4MD/e53v9PGxkZAZnyFjU+40YbS5zUYHv47aZzKRWGhDVvvk44H8nlqL8O1sWCd3FdFiJhKpazJOB4dHsH8/LxSqZR6vZ6q1aqFXVymFz6A4ZNeJykQPEB/Ym2S4c998z6f6QRHDZe3UXJDGOMPsXDS6/Dw0Kxst9s1emQyGfPoCN/n5+fN6yHhsrOzE7jVlvWHFSz/9srAl6FFIpPfCedbcQKB+KL1cPKPgRLm8Ik/xNJsNlWtVq0Ok7AUemQyGS0vLweSuaVSyfBzPKHNzU1rUM9zeDfr98dg2RvPWyhmabIDPN57I7QO08Unrn2lARiw90rH47Elq+EXvks0kc1mdfHiRcXjcWWzWQ2HQ4u8uO5qd3dX29vbJkeehz30El4r0YyvnEHOJy3n5NnhEkje43M5npep80X2oE2r1VK1WjWoC/2CHM3Ozmppacm8VfglmUwa3ILy3t/ft/X7iJJ5YbRYu+cp+NeXpJ40TlXAdMifmpqyQmlgAUlW1oHX50+1zM3NaWFhwfATWkdub2/r8ePHZt34PfhfoVCwxBRYJ5N/+vSpNjY29MUXX+izzz6zzfB1xP6qI2+ZfKLOY3qSAhbrLKNardr7APgRVjDqcAE52NX8/LwWFhYkyYD+fr+v7e1tPXr0yG46IEmSy+X0l3/5lyoUCiqXy1Zd4bPRP/74o9bX1/XFF1/oiy++CNAFpcGpI/Bu/3twN4ysvzhzEoHa2dmxkGtmZsb2AOV4eHhoeCjZdbBObnSWZFEBpXfQJZVKmRGvVCr6m7/5GxWLRS0vL1vG3Ifzjx490rNnz3Tv3j199NFHBoUAVcHX0vPm4GCdF2sB2AAAF1FJREFUvkQOfgGvhi6TGG28NzxdTxewTyIG+GU4HBpktbi4aEkiBHpvb08//fRT4IaZWCymcrlsdLl8+bLJEJFUNBrV48ePrXn//fv3AyVbvvoDHkTp84foEqPgD0dMMrj8018a6r1u9AvesFdyy8vLWl5e1ng8trwG0dcPP/wQKB/s9/vKZrP6q7/6K83NzenSpUsmZ94AQ5dPP/1Un3zySSC5iMwAJ6ELuVjV55y4lbzVapne+VlVEAw20IPsYcvoGQeFDGP4jUFJQgCEE2yUWw7oHxCNRu3qcCw2jbV5ng+1sVBh8DvsAfr6wEnCSenY+/F0Yf2eLjCSrx/1B0j88KfZAPa5PmV+fj5AF95LYgk4hw5ivNcraeYT3g8+47PR0HKSU03ScfQDvMP74AtoB9zka2t9D2XvAWKIvYeMwl5YWFChULCLONlHKm12dnYCdPH77pOo0CG8J55enm6+7GsSuoSz4d5Lwvv2NcaeLp5voR/CT9ljOp3W4uKiFhYWrCTRH8ohsURUwIWvvMuv3++Bx8b5HTTDk0dZTzI87u5pjBz5iAA5Ql45LOH3cjweG0REQhz9sry8rHK5bPXUwHrRaFR7e3s6PDw0upBzgVeYm4cq8XL5mS9/9dUzrxqnah4wHV9gT2KJazZgEAB1qhq4/iOTydjlfb1eT9euXdOHH35ohCmXy3rjjTdUKBT02muvBUqA8Jrv3bunJ0+eaGNjQ7u7u2o0GgF8ztd1wrz+eh8qNwhRyfpyYuusxPJ08Z4WWBBVIazNH16BRjQZgYGw3NevX9eHH35o7+C0E3edwYCj0UjNZlNHR0e6f/++tV4Ed+XqFrxa4A6EPHwNDvMg6UnISsnSJKG25xeUBF4uIRuKI5fLWbTgb/2APjzr2rVr+tM//VOlUinlcjmVy2W9/vrrmpub0xtvvBFodE9lzMcff6yffvpJa2trZrCTyaR5IzgBJHSk5+VeJDfhKW/IvIEP43yvGjwrzGftdttyJ0QO6XTaPD1KrVAkzGN6elpXr17VL37xC3N0FhcXdfv2bc3Nzen27duBC2fp//ub3/zGoqXd3V3V63XjKyqEOPiA0ud3J9X+szbK3067a++kQdKWslCMKFUhOGFEmsgyh7ugTyaTMSjm8uXL+vM//3NzApeXl/XOO+8on8/r8uXLpstGo5H1/b13756ePn2qP/zhD1pfX1e73bbCAfjFw4bj8diOKjNwJpGjSCRiN1Gf5HAZb5xGICxQuM6PiYU9Kb5zksXASlBWA+NUKhVduHDBFLUkK3Uh4YBl4ooZhPm0OYcH8IAXwtOyl6+ii1f2PAcG8R61/1x4Pt7jzGazBk2Mx2MtLCwYxgmWfnh4qF6vp4ODA7VaLW1tbdn1TLVa7QX8LPx/hsfrT/p82FM76wgfUuCdPmz1iuik9/iKiVjs+UnKSqWidDptuO/ly5ft/i9PF/iFi0nhFxIn4fed9G8fZqOcPPbocfOzDg99ScEWmD6LzjtwcsJ08bKVy+WsmgEFzFVeOBb04IZfNjY2tLq6GpAjjyufxDN+rsBqXtY9Nv1z8ike2mAfeK+PNKnXfhlf8pnZ2VktLi7a/xcXF3Xx4kXrfQG/9Pv9M8kRA7p7vN2PcKTgDfZp+uVUBbyxsWHKgVNaCBn4CrVybFC9XjfPjmtQ8Pjy+bxu376tSqViXg4eEfWb7XZb33zzjQ4ODvTFF19ob29POzs7ajabxrDgLT45SFE3Fg78hZD9pDCUC0Uh1FnHzs6OJQI89CIdd3g6PDy0wygkkygUp10etcKpVEqvvfaa1b5SB+lrL6ELJwR3d3et5pd3U0DO+sgwE8qPx2M7wIKRaDabqtVqgUMjXCI5qSKmgf/s7Kztiz/04SMWvN9qtarp6WnV63WLmjgFlUqldPfuXS0uLlrICByBp0T7wL29PX3yySfa3t62CgrCe7BnhIFSLjBwSVYC6ZOcVG74VoSxWOyFI8Bn4ZdIJGKNhBhEbsgKITV70Gg0VKvVLHIiipuZmdGdO3e0vLxsmKOPusL88vHHH5sT02w2DZMEe/aVMewXdGk2mwFogaiIyInvsL5J+GVlZSWATUvHjhz7hBwnk0n1+/0X+GVvb8+85fn5eb3zzju6ePGi9YKAbtFo1PjgwYMHRheS1mC6RLK+ibt03NwLg8OFD+Qewl31OBUbj8dVLBZ/ngfsG5ZIx7gvoYIH6FGOhPgoA4qbUSqzs7NGGM59Y1VoU7m9va2dnR09evRI29vbAcUPg1JC5asQvMIAQ0LRUBLm8VqINanVJqEWLkgnqeKP2EIbiusxHiRdEPhcLhfAtvwhCvq+bm5uWhKTUjP2A4YllEQB+2SgDye9p+Wz0V7RePzwrHSJRqOmoHwGGZzQe1q8m/Krdrtte4QA0VqSNfpoA37Z2NjQ1taWvvvuO21ubko69naBDFCuGGg8cn8qDNqhmIEkwl4v9JuELjgH4ZpZX2kQ9jahCxAChjQWi1lfB3IpfB7j32g0jF9++OEHbW5uBrqweSjFV3ZAFxQwxgG6AGVJx43K/ZHfSRQwRtDnIZAjnu3liHdxyAaFSriPskMphtsj9Ho9o8vW1pa+//57bWxs2F76NbKmcGUQsg7vgkEzT8r1cBSh4cscvFMV8NLSUgDDYyG+AQ/YWrPZNGXQ7/f12WefqVgsGr47Ho+Vz+eN4enoj5Ku1+t69OiRnVhpt9tWZ0ubPZqM4IVAmKmpKWvY0mg07FQaTDUYDJTNZlUoFIyp8aQGg4FqtdpEYeWFCxfM+/eWkQQAmyc9v46GZ3c6HX399ddaX1/XzZs3lc/nTRhJZnS7XTWbTTvtV6vVjC6PHz+2e85820nmEYlEAt4ZNZC82x9IgcnB2DCc4JAw0CR04XJV8HsMBMbX90o+ODgwXur1erp3754eP36s999/X6VSSeVyWdls1iCier1u195HIhFVq1V99dVXqtfrRheMNacL8SKZA8aW5k3j8TgwD5SjF2hw2kQioXw+b8m9SbDOYrEo6fiSWRStbx+JsajX68bT3W5Xn3zyiebn5/Xuu+/aYQHwc0lWckX1yv7+vh4+fGjyxE0r3W7X+AFjJynQmwNFKB3jsuDyvpoI77nT6Vht988Z8AuOBmsi8eq97Gq1avql1+vp888/15MnT6x+98r/acID7ZrNpprNphmmarVqTayePHmiVqul3d1dDQYDZTIZxeNxuycvEolYxU273VYymdTc3JxGo5GVpoWjFyAx8lbopH6/b/XEJ41TFTAdf3gplshXGaAY2VAItrKyYh2YxuOxNUTB6nJ5XrfbtWOV9+/fV71e18bGRqDxCkrBn+rxYSyKxONQWHkUAQc+KFkhCdfpdAL1oWcZuVzOQhU8CvAwhIkwBLrgcWxubqrdbmt+fl6RyPMWeDTslmSKF6O0ubmp3//+92a5+/2+VUNwJNyf5kKAmA/7QdMQKdjUmpaEJLD4jj+aedZBpzXe5b+LcNMLgWOoJNGePXumWq2mCxcumCFhfijCnZ0de97a2pru3btnR9vxVnlPIpFQs9m0aMVXevBceJEjyHgukgLwEYaLe8ngobMOTmqRKII2HBBgoNhYx2AwMDlaWloy7Ddc2bC3t2cGdH19XZ9++qlqtZodPMHwwS+E29DFlzX6SxhQeJHI81ukwzfWUFKIwT7t5odX8YvHoYGK4GFkAt4cj8fW6a1UKmk0GtlZARyGbrdrJx/pA8GNx9DLV1JRNkkimsQie8V8iL45DOT7nfgLCTzchVN40nhlMx4IBXN4DA93HYaRZK4+hwo+/vhjpVIpPXz40G4o4Dl4Awg7DOrrGhEaLrnEAvqOYbFYzLLJ3jMmkTI1NWWdkCj6j0ajtr5yuTwR43i6cEAEJQgs4RUhP+coNnh0JpPRgwcPlEqlrDbUJwoxVP60FrSORCJmkKanp5XP5y15CUYfjR5f6OmPa3q6oFBGo9ELDZYWFhYmCilhMjwiemSwn/CHdFxZw/dIjPz2t78NHMpBARJWS7JECieoMHweVuEyR25OwejyObBMbkzAuSAiAEcdjUbK5/PGQ9FoVFeuXJnIMLFmyuhocEM04B0I7w2CVQ+HQ927d0+pVMqOJVM/jvOD4qFxEXTHEfEJRM+3KGI+g9IjQvCN6jGeg8HA+nKjwBOJhObm5ibiF6AZqgV8fxfoSxWRpw/Hrdvttj755BOl02l9++23KhQKFiWjT6Alt2R4OMAbsn6/bzkp+IGcQyQSUaPRMKOJsxOJRCyKHQwG2traMoeR52YyGd24ceOlNDhVAQNEc/UNQs8CvDX0WWSSTs1mU2tra5KeH8xIpVJ275L3SmAOsFk21XstvugZDwSccjx+sVkGIDkChwKmbpSz4IlEQsVi8WcxTjabVSwWszBXOi5h8hUghIdEE5FIRAcHB4rFYoZXUdMbDgdJUiAYvpQJpUGILymgaIACsM7sKTQmesFLohyNeRSLxYmSk8wRzy4SiVg45jPnUvCkHKEeHc7G47EKhYJ55hzn5rv+YAEek1fCQGXATP4GW46st1otTU09P9YajUbt9CBllURo1B/j2czMzNjBiLMOlC1etsdYB4OBJalR1D7TTkhLiRfHzqGLPzDiHZfxeGy5BPhGei6feMKtVssikXAlRhjH9woYWSwWi+YRJhIJzc/PT8QvGE9wW++V8k5fSgpvAnlSwcAz0um03QPocxnwi6/I4tkYHQ5tZTIZ6zJH4pcKGyKBSOS4UT2XIayvr6tarVr7XXg9kUhYD5uTxqkK2DfzxqsCL+z3+4F6T19byiZKCiSUcM3xbFBaPgkgHdeTgjuDlWEAqEDwpVw+uSHJ6gbZADYXgjMP782fdfiQjnpLsG88Lyw3ClY6bsHoy26glb+ZwR9a8GEzdIMJgSJYE5g874K23gNCmfjnsod4GWTjMWpnHb6GFH7Ay/Ynv3yVBgMaUUXCGlCajLAihn6SLBKBT/mDosazAuMLJ7W8UkQJsdeszx8QmoQuGFTCfEJ2DKQvs/JF/tDL09JHWWHB9ok0aMkJOo9x831fN040xZyl53JERh/ly+fpFsfx6vX19TPTRFKgRwNzQdm2Wi2rjvIRAgoZHqEmud/vW6UUOgda+nV7Oce7R7+QfPMVK/Clb/LO3CWZ1wxEircMjs17flYVBC9BgNlQsFusLp4cHpdXHJJsgygu55z+4eGheWssHi+WiRNKceSXRBEYHgdCCHvBrthcCE+4BROHDylMMiibAbyXjjEnn4zEq6Qcjtpm6IIhQNiTyaSFkNBNUkABgylxvt3XHFMwTpaYZKVnPqyzD/dQUrTylGTlW5OE2hgRoCLpuTLzUAHGPMyQ3jNmgK8jhHiOvk2kr2rxEBk3XngDg/HyBg8+IbTkhmkUP8YIulD2NYmnFzZMPtHlqxL8PrFmH84STgMreWMG9OeVBMbY/9yvhXkBYxBNgrUyp36/r52dHYsaoDmJvWQyqVarpWfPnk2USwEKZB7QdW9vzzxOf5rNe8JEuChp75nTkB0aYTChLVGQh1Oy2az1TvYHgaTnckjehx7COD/Qg4SxLx2kQIFKjZPGK5vx+MyoFxIW1ev1AuVBMAvfoRQkHJJ7rxOLjcLHC/Seoy9xImSQjk/nea9WUgDfgvgkrrB4eNWTeDPMTzpuyuyBesrJfLjDGqRjowbTMQ8vjP6EnvcMYDLmDRMRwhKuYgA8ZMHJL6IUnyRjHiglIgboetbBM8Pleb4bFPP3tbXSi5g5nie4NorYKxIMNZESwkAIjnGnSsBXQxBSE+WRmPKli5JMODH2fG9SnsFzh/YoOowe8A2ennR8g4vnJYQeXvC8xtyRAejrISAgD3IpePsoQ09DZAcl6RUZSW3oQd/dSRQwn8Uo4rXC7yh57+l7mEtSwFPleTzHQwzS8QlWZMBHp/AlkTX6xhs/Ikg+7+uj+Te17N4xoJHUSeNMl3L6MJdQBuGgnhWPA+XgwzQ2cHp62k6beNwYJikUChoOn7d59K0SWSz4LR3vU6mUhQaUYXHzxtrampXJeKbJ5/Mql8umxL0xOetAWTYaDYMwpOdWkVI3vEvWiBIpFAoBj84n1CjGJ8nC/GZnZ9Xv97W3t6der2deLTWiCwsL1hWt2WzacWLpGO8ikUTpnz8UgadMht33nZ2kDI0SHtYjBc/nHx0dmRdFUhDvAOEAW6MMjEJ5sFnWFI/HVSqVNBwOtb6+bjgk9BwMBsrlcnasmgNB/lZlaCvJTkV5bDAWOz5ZRdkS8NIkHrAU7JAXjuyQI08X78h474mTkTs7O2o0GhYCM+irAl3gqUgkYpBHJpMxL5HkdqlUspJTlGosdnxrs4fbxuOxcrmcVfJgOLwsnGXgyRPR8Ix8Pq98Pm/8QqWOr5KYm5szA+vpAwYMj3h8l4ig0WgY9IBCPjw8VDab1dzcnHVcZD/CPCdJT548sbzNYDCwo9E02+IKMQ46/awqCJgQpoT4hATADN57wZphAbAcCANKw3vBvMN7pJ6oCIxv2ILnzXvx/Hx1BkkV70mDZ3lPZNI+pv6cuLeC3ljhqeBNgn9zSgxvyAt7NBo1IcBDCic8mbf3CsLf5d+s2T8fI+qhIObXbDYtdMILmBSCQLGhBH0yxfMF/W+ZL/uKt1oqlayPNM9jLvBMPB43Y9jtdgPK1UdkYU/SQ05UEGCU8A7r9bra7bYymcwLuCfwzlmHh0H4HnuIUoFONA/f398PeMIIsL9BBpowN+m5ws5mswHc1t+tJsnKrKg3R/bgmWg0ag5NOEyvVqum+JFBIDn4cxJ+gRbIcLfbtePU0Mp7sr7qKhKJmJPAjek4axg8+I0oL1waSaK43++rWCwa/ev1emBNRAs4eH4u8DrJXowotAkbUT9e2YwH74iwp9PpaH5+PrDJLIo+o2CN1MlFIhHduHFDV/5P+Q4N3D2MQP9gPAPwHn5PeF8sFrW/v2/eCgcGuEetVCrZXLvdrnK5nNLptBqNhmGrgOWEO74t3VlGpVLRYDDQs2fP7JmSrGG8T9TEYjHrZxGPx60XKQkA8DY2mPInf0QbY+IFjVAymUzalT+DwcAK+SmvouoBRsHTvnbtmubm5vT48WOrF93f39fCwoJmZ2fV6/WsUP2sg14WKF6warxGEkjpdFrFYlH1et3utyP05sTW3bt3de3aNWvrR1SBspienlY2m7UGSChLf6zVCzDCjnCyLrwh+DifzyuTyWh9fV3Pnj0zgSJnwb5NkrSFFwjZOfaMUSQkpmH/+vq6PvvsM/X7fev3ADxVLpd1+fJlHR4eam9vT6PRyGpk8dguXbqkZrOp3/zmN4Ej3ij8bDarbDZrWCs8iGLp9Xp22QHe+N27d1Uul/XRRx/p6dOnVjlTLBZVqVQ0Ho8t8X3W4S82oB6/2+1qbm5OFy9eNA8Y5Ua1jE/ar62tqd/v68MPP7RmO1NTx20NwnI+GBz3oob+3B/3+uuvW0MwyiLRebVaTZlMRteuXQvUBCPr3NVYKpVUKBQM4kEX/SwPOOxJeA8ExYBn4Y/dSsc4DH97rxCPIFxW5LEuz+B8H0uG5833vXfoPQpgDo/joMz4+SQMw8BbCFtB/yzvcUnHCSUpeHGhTw6EIZmXwSJ4+eHv8TtPl3DFhU+MEnF4LBahC6/vLAMB53t+Li/bU/98n2Qh4vF4rF9LeF/DEQOfCf/b84qnF3+Yp59L+DtEFWcdnq/993wZp3QcQcVix8eEMdJ81icl/dr985ETX1HjZTiMbfJuz08+QpSCNzswt3DFE9j7WQd75TFsX73gPXP/mbDc41D505Z+PSfJGO/10bLH2KEfPIcD5J/h6e4dRo+d8+6XOXiRn6OAzsf5OB/n43z834/JUrnn43ycj/NxPv5fG+cK+Hycj/NxPv5I41wBn4/zcT7Oxx9pnCvg83E+zsf5+CONcwV8Ps7H+Tgff6RxroDPx/k4H+fjjzT+H9cply6jPWTBAAAAAElFTkSuQmCC",
      "text/plain": [
       "<Figure size 360x144 with 5 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "manager = GradientInversionAttackServerManager(\n",
    "    (1, 28, 28),\n",
    "    num_trial_per_communication=5,\n",
    "    log_interval=0,\n",
    "    num_iteration=1000,\n",
    "    optimizer_class=torch.optim.SGD,\n",
    "    distancename=\"l2\",\n",
    "    optimize_label=False,\n",
    "    lm_reg_coef=0.01,\n",
    "    lr=1.0,\n",
    ")\n",
    "CPLFedAVGServer = manager.attach(FedAVGServer)\n",
    "\n",
    "client = FedAVGClient(\n",
    "    LeNet(channel=channel, hideen=hidden, num_classes=num_classes), lr=1.0\n",
    ")\n",
    "server = CPLFedAVGServer(\n",
    "    [client], LeNet(channel=channel, hideen=hidden, num_classes=num_classes), lr=1.0\n",
    ")\n",
    "\n",
    "local_dataloaders = [DataLoader(TensorDataset(x, y))]\n",
    "local_optimizers = [optim.SGD(client.parameters(), lr=1.0)]\n",
    "\n",
    "api = FedAVGAPI(\n",
    "    server,\n",
    "    [client],\n",
    "    criterion,\n",
    "    local_optimizers,\n",
    "    local_dataloaders,\n",
    "    num_communication=1,\n",
    "    local_epoch=1,\n",
    "    use_gradients=True,\n",
    ")\n",
    "\n",
    "api.run()\n",
    "\n",
    "fig = plt.figure(figsize=(5, 2))\n",
    "for s, result in enumerate(server.attack_results[0]):\n",
    "    ax = fig.add_subplot(1, len(server.attack_results[0]), s + 1)\n",
    "    ax.imshow(result[0].cpu().detach().numpy()[0][0], cmap=\"gray\")\n",
    "    ax.axis(\"off\")\n",
    "plt.tight_layout()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "e-TYM84K_9wn"
   },
   "source": [
    "## Reconstruct Batched Data\n",
    "\n",
    "Second, we simulate the situatino with larger batch size. We try to recover the bellow three images."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 89
    },
    "id": "CIrFdMAF-4LX",
    "outputId": "2436568f-f281-4b1a-8629-a84b4872b5ad"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 216x144 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "batch_size = 3\n",
    "x_batch = xs[:batch_size]\n",
    "y_batch = ys[:batch_size]\n",
    "\n",
    "fig = plt.figure(figsize=(3, 2))\n",
    "for bi in range(batch_size):\n",
    "    ax = fig.add_subplot(1, batch_size, bi + 1)\n",
    "    ax.imshow(x_batch[bi].detach().numpy()[0], cmap=\"gray\")\n",
    "    ax.axis(\"off\")\n",
    "plt.tight_layout()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "Z8Sh_1Fn_VeG"
   },
   "source": [
    "### GradInversion\n",
    "\n",
    "https://arxiv.org/abs/2104.07586\n",
    "\n",
    "- distance metric = L2 norm\n",
    "- analytically estimate labels from the gradients\n",
    "- reguralization: total-variance, l2, bn, and group-consistency"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 89
    },
    "id": "5Dnp5tQoA1qm",
    "outputId": "07e5fe9c-eedd-4496-8e56-40538f072e5c"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 216x144 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from aijack.attack.inversion import GradientInversion_Attack\n",
    "\n",
    "net = LeNet(channel=channel, hideen=hidden, num_classes=num_classes)\n",
    "pred = net(x_batch)\n",
    "loss = criterion(pred, y_batch)\n",
    "received_gradients = torch.autograd.grad(loss, net.parameters())\n",
    "received_gradients = [cg.detach() for cg in received_gradients]\n",
    "\n",
    "gradinversion = GradientInversion_Attack(\n",
    "    net,\n",
    "    (1, 28, 28),\n",
    "    num_iteration=1000,\n",
    "    lr=1e2,\n",
    "    log_interval=0,\n",
    "    optimizer_class=torch.optim.SGD,\n",
    "    distancename=\"l2\",\n",
    "    optimize_label=False,\n",
    "    bn_reg_layers=[net.body[1], net.body[4], net.body[7]],\n",
    "    group_num=3,\n",
    "    tv_reg_coef=0.00,\n",
    "    l2_reg_coef=0.0001,\n",
    "    bn_reg_coef=0.001,\n",
    "    gc_reg_coef=0.001,\n",
    ")\n",
    "\n",
    "result = gradinversion.group_attack(received_gradients, batch_size=batch_size)\n",
    "\n",
    "fig = plt.figure(figsize=(3, 2))\n",
    "for bid in range(batch_size):\n",
    "    ax1 = fig.add_subplot(1, batch_size, bid + 1)\n",
    "    ax1.imshow((sum(result[0]) / len(result[0])).detach().numpy()[bid][0], cmap=\"gray\")\n",
    "    ax1.axis(\"off\")\n",
    "plt.tight_layout()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "background_save": true
    },
    "id": "cjSDtkOYC4AF"
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "provenance": []
  },
  "gpuClass": "standard",
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}