docs/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."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "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": 2,
   "metadata": {
    "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": 3,
   "metadata": {
    "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": 4,
   "metadata": {
    "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": 5,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 85
    },
    "id": "z-dkdI2E5gE_",
    "outputId": "ce242131-adb8-4e00-cd66-de817ccbf70f"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 72x72 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "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": 6,
   "metadata": {
    "id": "UA3YUHqi6nq2"
   },
   "outputs": [],
   "source": [
    "criterion = nn.CrossEntropyLoss()"
   ]
  },
  {
   "attachments": {},
   "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": 9,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 108
    },
    "id": "4BAe6aEu5odr",
    "outputId": "0f637df8-817d-4179-de31-0441045ddf08"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "communication 0, epoch 0: client-1 2.641099691390991\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=100,\n",
    "    distancename=\"l2\",\n",
    "    device=device,\n",
    "    gradinvattack_kwargs={\"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": 10,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 162
    },
    "id": "PJAu_poY6q7L",
    "outputId": "cc804e87-34b0-47b2-b765-ee63af5142f9"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "communication 0, epoch 0: client-1 2.371312141418457\n",
      "iter=72: loss did not improve in the last 50 rounds.\n",
      "iter=75: 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",
    "    gradinvattack_kwargs={\"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": 11,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 108
    },
    "id": "bv0qW14888Kh",
    "outputId": "dc077c7c-847b-4826-abc6-1590c1cdc4a1"
   },
   "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",
    "    gradinvattack_kwargs={\"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": 36,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 108
    },
    "id": "Lz-UJfvR9MDZ",
    "outputId": "69b1cc96-e80f-4fac-b641-b9885fe4bd1b"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "communication 0, epoch 0: client-1 2.5491995811462402\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",
    "    lm_reg_coef=0.01,\n",
    "    gradinvattack_kwargs={\"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": 23,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 89
    },
    "id": "CIrFdMAF-4LX",
    "outputId": "a65a3711-2f7a-4045-95e4-e8f2592d5b69"
   },
   "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": 24,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 89
    },
    "id": "5Dnp5tQoA1qm",
    "outputId": "d538b96e-e2a6-46e0-b9b6-f36a34a07640"
   },
   "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": {
    "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
}