GDP_NN_tutorial.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Differentially Private Neural Network Tutorial with GDP.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "toc_visible": true,
      "authorship_tag": "ABX9TyNVrfxRUe9mQx+55O8KGzwv",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU",
    "widgets": {
      "application/vnd.jupyter.widget-state+json": {
        "3b0c428edd0b41fab0a8567e2b98a241": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_7eba259865cf468cb4a0104989ac0ac9",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_ddb9e60f1d7c4c95b94b8c0a6f7a1912",
              "IPY_MODEL_bdea42859db84ce7b64ad70fa6f42ef0"
            ]
          }
        },
        "7eba259865cf468cb4a0104989ac0ac9": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "ddb9e60f1d7c4c95b94b8c0a6f7a1912": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_33ade62d1377499f80bb176184e62421",
            "_dom_classes": [],
            "description": "",
            "_model_name": "FloatProgressModel",
            "bar_style": "success",
            "max": 9912422,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 9912422,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_eee3ecc05d3b489ca9cadcbebe5ce6f7"
          }
        },
        "bdea42859db84ce7b64ad70fa6f42ef0": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_7da234722a5045c186dd36b92cdfda1b",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 9913344/? [00:20&lt;00:00, 481297.25it/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_7f3ad14f06214429bd0033b9cb545c94"
          }
        },
        "33ade62d1377499f80bb176184e62421": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "eee3ecc05d3b489ca9cadcbebe5ce6f7": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "7da234722a5045c186dd36b92cdfda1b": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "7f3ad14f06214429bd0033b9cb545c94": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "c2b5ba655f184aa5b1d5a8bb250e08f0": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_9c1c7adcdbef40808e6bcf6a83c87256",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_cc6e324133154e99885566b30459a1f8",
              "IPY_MODEL_84c42f1692184555814c02ca52b3f57a"
            ]
          }
        },
        "9c1c7adcdbef40808e6bcf6a83c87256": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "cc6e324133154e99885566b30459a1f8": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_ac75d52193ef49f5a15a0fb24c1b6ea3",
            "_dom_classes": [],
            "description": "",
            "_model_name": "FloatProgressModel",
            "bar_style": "success",
            "max": 28881,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 28881,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_c0a586dbf5cc403aa3272a636ce2a9d1"
          }
        },
        "84c42f1692184555814c02ca52b3f57a": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_1791698f0319486cae41d9b0a2783948",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 29696/? [00:02&lt;00:00, 12482.92it/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_5a95fba0e9c246c3bdad24c4016588bf"
          }
        },
        "ac75d52193ef49f5a15a0fb24c1b6ea3": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "c0a586dbf5cc403aa3272a636ce2a9d1": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "1791698f0319486cae41d9b0a2783948": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "5a95fba0e9c246c3bdad24c4016588bf": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "83d9a715473c4cdaa6759ab2d100352e": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_6f91954fdad4479e87dcc785b792167b",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_f73c90d2c2474260be70971c4dc78f1a",
              "IPY_MODEL_c7cbebc2c7a0484a871f8b49bec0cd8c"
            ]
          }
        },
        "6f91954fdad4479e87dcc785b792167b": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "f73c90d2c2474260be70971c4dc78f1a": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_e668a7dba5014fd1b3d5f92eb60fc9c7",
            "_dom_classes": [],
            "description": "",
            "_model_name": "FloatProgressModel",
            "bar_style": "success",
            "max": 1648877,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 1648877,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_112403216e964cb6a4c57cd94fd2c077"
          }
        },
        "c7cbebc2c7a0484a871f8b49bec0cd8c": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_6756211697eb4d95a40caec7f0ccda2b",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 1649664/? [00:01&lt;00:00, 1471610.62it/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_b75888b1e5074d6499c0f6f57bc39f3b"
          }
        },
        "e668a7dba5014fd1b3d5f92eb60fc9c7": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "112403216e964cb6a4c57cd94fd2c077": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "6756211697eb4d95a40caec7f0ccda2b": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "b75888b1e5074d6499c0f6f57bc39f3b": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "f56c46fd31e846fea05b864f7e22fa04": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_b5977a5eedac488f930516d7414b0385",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_1187ee7db99044248e5bbb1f625db20d",
              "IPY_MODEL_434f7c8abc034b5ea29877447c148e30"
            ]
          }
        },
        "b5977a5eedac488f930516d7414b0385": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "1187ee7db99044248e5bbb1f625db20d": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_34387d55a6c74701b2bffda7bc86f5ee",
            "_dom_classes": [],
            "description": "",
            "_model_name": "FloatProgressModel",
            "bar_style": "success",
            "max": 4542,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 4542,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_b6be9a3ae3fe48bd86a8555a89048901"
          }
        },
        "434f7c8abc034b5ea29877447c148e30": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_0870bc3ca5ca44c98ef398d7675bd6b3",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 5120/? [00:15&lt;00:00, 320.89it/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_f079d864cd554548a8bed7aa1573a5f2"
          }
        },
        "34387d55a6c74701b2bffda7bc86f5ee": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "b6be9a3ae3fe48bd86a8555a89048901": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "0870bc3ca5ca44c98ef398d7675bd6b3": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "f079d864cd554548a8bed7aa1573a5f2": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        }
      }
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/woodyx218/Deep-Learning-with-GDP-Pytorch/blob/master/GDP_NN_tutorial.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "temGVWcunZXf"
      },
      "source": [
        "# Differentailly Private Deep Learning accounted by Gaussian differential privacy\n",
        "\n",
        "### This is a tutorial of how to train deep neural networks with differential privacy, a mathematically rigorous privacy definition that defends against privacy leakage, e.g. the [membership inference attack](https://github.com/tensorflow/privacy/tree/master/tensorflow_privacy/privacy/membership_inference_attack).\n",
        "\n",
        "### Our tutorial is based on the Pytorch library, [**Opacus**](https://github.com/pytorch/opacus). First we install Opacus."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "r9qUt7WqoX6L",
        "outputId": "c2afe6aa-ea16-4dba-e293-d46ee4942a02"
      },
      "source": [
        "!pip install opacus"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Collecting opacus\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/68/c4/36f07b54413e607f07858d5b668379991948f0de9979185a895fee9c2208/opacus-0.13.0-py3-none-any.whl (102kB)\n",
            "\r\u001b[K     |███▏                            | 10kB 9.6MB/s eta 0:00:01\r\u001b[K     |██████▍                         | 20kB 14.9MB/s eta 0:00:01\r\u001b[K     |█████████▋                      | 30kB 19.4MB/s eta 0:00:01\r\u001b[K     |████████████▉                   | 40kB 19.4MB/s eta 0:00:01\r\u001b[K     |████████████████                | 51kB 11.9MB/s eta 0:00:01\r\u001b[K     |███████████████████▎            | 61kB 13.1MB/s eta 0:00:01\r\u001b[K     |██████████████████████▌         | 71kB 10.6MB/s eta 0:00:01\r\u001b[K     |█████████████████████████▊      | 81kB 11.3MB/s eta 0:00:01\r\u001b[K     |████████████████████████████▉   | 92kB 12.1MB/s eta 0:00:01\r\u001b[K     |████████████████████████████████| 102kB 5.8MB/s \n",
            "\u001b[?25hCollecting requests>=2.25.1\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/29/c1/24814557f1d22c56d50280771a17307e6bf87b70727d975fd6b2ce6b014a/requests-2.25.1-py2.py3-none-any.whl (61kB)\n",
            "\r\u001b[K     |█████▍                          | 10kB 20.6MB/s eta 0:00:01\r\u001b[K     |██████████▊                     | 20kB 26.0MB/s eta 0:00:01\r\u001b[K     |████████████████                | 30kB 31.2MB/s eta 0:00:01\r\u001b[K     |█████████████████████▍          | 40kB 23.6MB/s eta 0:00:01\r\u001b[K     |██████████████████████████▊     | 51kB 20.1MB/s eta 0:00:01\r\u001b[K     |████████████████████████████████| 61kB 6.7MB/s \n",
            "\u001b[?25hRequirement already satisfied: tqdm>=4.40 in /usr/local/lib/python3.7/dist-packages (from opacus) (4.41.1)\n",
            "Requirement already satisfied: torchvision>=0.9 in /usr/local/lib/python3.7/dist-packages (from opacus) (0.9.1+cu101)\n",
            "Requirement already satisfied: numpy>=1.15 in /usr/local/lib/python3.7/dist-packages (from opacus) (1.19.5)\n",
            "Requirement already satisfied: scipy>=1.2 in /usr/local/lib/python3.7/dist-packages (from opacus) (1.4.1)\n",
            "Requirement already satisfied: torch>=1.3 in /usr/local/lib/python3.7/dist-packages (from opacus) (1.8.1+cu101)\n",
            "Requirement already satisfied: idna<3,>=2.5 in /usr/local/lib/python3.7/dist-packages (from requests>=2.25.1->opacus) (2.10)\n",
            "Requirement already satisfied: urllib3<1.27,>=1.21.1 in /usr/local/lib/python3.7/dist-packages (from requests>=2.25.1->opacus) (1.24.3)\n",
            "Requirement already satisfied: certifi>=2017.4.17 in /usr/local/lib/python3.7/dist-packages (from requests>=2.25.1->opacus) (2020.12.5)\n",
            "Requirement already satisfied: chardet<5,>=3.0.2 in /usr/local/lib/python3.7/dist-packages (from requests>=2.25.1->opacus) (3.0.4)\n",
            "Requirement already satisfied: pillow>=4.1.1 in /usr/local/lib/python3.7/dist-packages (from torchvision>=0.9->opacus) (7.1.2)\n",
            "Requirement already satisfied: typing-extensions in /usr/local/lib/python3.7/dist-packages (from torch>=1.3->opacus) (3.7.4.3)\n",
            "\u001b[31mERROR: google-colab 1.0.0 has requirement requests~=2.23.0, but you'll have requests 2.25.1 which is incompatible.\u001b[0m\n",
            "\u001b[31mERROR: datascience 0.10.6 has requirement folium==0.2.1, but you'll have folium 0.8.3 which is incompatible.\u001b[0m\n",
            "Installing collected packages: requests, opacus\n",
            "  Found existing installation: requests 2.23.0\n",
            "    Uninstalling requests-2.23.0:\n",
            "      Successfully uninstalled requests-2.23.0\n",
            "Successfully installed opacus-0.13.0 requests-2.25.1\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ZdP2WEtQx9ls"
      },
      "source": [
        "### We also need the privacy accountants including our GDP accountant"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "yZqPH9b5x8F-",
        "outputId": "4d80048e-b4c2-4a2a-ec03-ad7a60cbf0d2"
      },
      "source": [
        "!git clone https://github.com/woodyx218/Deep-Learning-with-GDP-Pytorch.git"
      ],
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Cloning into 'Deep-Learning-with-GDP-Pytorch'...\n",
            "remote: Enumerating objects: 33, done.\u001b[K\n",
            "remote: Counting objects: 100% (33/33), done.\u001b[K\n",
            "remote: Compressing objects: 100% (31/31), done.\u001b[K\n",
            "remote: Total 33 (delta 12), reused 0 (delta 0), pack-reused 0\u001b[K\n",
            "Unpacking objects: 100% (33/33), done.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "eBJdJFTuz5PX",
        "outputId": "c235a51e-e6b4-425f-e117-ba237c02be45"
      },
      "source": [
        "cd /content/Deep-Learning-with-GDP-Pytorch"
      ],
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/content/Deep-Learning-with-GDP-Pytorch\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "NCJ3nkLHondV"
      },
      "source": [
        "### Import necessary packages."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3L6X2njWomvG"
      },
      "source": [
        "import numpy as np\n",
        "import torch\n",
        "import torch.nn as nn\n",
        "import torch.nn.functional as F\n",
        "import torch.optim as optim\n",
        "from opacus import PrivacyEngine\n",
        "from torchvision import datasets, transforms\n",
        "from tqdm import tqdm\n",
        "from torch.utils.data import DataLoader"
      ],
      "execution_count": 7,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6iSmvINfowkc"
      },
      "source": [
        "### We use MNIST here, similar to \n",
        "https://github.com/pytorch/opacus/blob/master/examples/mnist.py"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 823,
          "referenced_widgets": [
            "3b0c428edd0b41fab0a8567e2b98a241",
            "7eba259865cf468cb4a0104989ac0ac9",
            "ddb9e60f1d7c4c95b94b8c0a6f7a1912",
            "bdea42859db84ce7b64ad70fa6f42ef0",
            "33ade62d1377499f80bb176184e62421",
            "eee3ecc05d3b489ca9cadcbebe5ce6f7",
            "7da234722a5045c186dd36b92cdfda1b",
            "7f3ad14f06214429bd0033b9cb545c94",
            "c2b5ba655f184aa5b1d5a8bb250e08f0",
            "9c1c7adcdbef40808e6bcf6a83c87256",
            "cc6e324133154e99885566b30459a1f8",
            "84c42f1692184555814c02ca52b3f57a",
            "ac75d52193ef49f5a15a0fb24c1b6ea3",
            "c0a586dbf5cc403aa3272a636ce2a9d1",
            "1791698f0319486cae41d9b0a2783948",
            "5a95fba0e9c246c3bdad24c4016588bf",
            "83d9a715473c4cdaa6759ab2d100352e",
            "6f91954fdad4479e87dcc785b792167b",
            "f73c90d2c2474260be70971c4dc78f1a",
            "c7cbebc2c7a0484a871f8b49bec0cd8c",
            "e668a7dba5014fd1b3d5f92eb60fc9c7",
            "112403216e964cb6a4c57cd94fd2c077",
            "6756211697eb4d95a40caec7f0ccda2b",
            "b75888b1e5074d6499c0f6f57bc39f3b",
            "f56c46fd31e846fea05b864f7e22fa04",
            "b5977a5eedac488f930516d7414b0385",
            "1187ee7db99044248e5bbb1f625db20d",
            "434f7c8abc034b5ea29877447c148e30",
            "34387d55a6c74701b2bffda7bc86f5ee",
            "b6be9a3ae3fe48bd86a8555a89048901",
            "0870bc3ca5ca44c98ef398d7675bd6b3",
            "f079d864cd554548a8bed7aa1573a5f2"
          ]
        },
        "id": "bhVolH-PowGd",
        "outputId": "3a3e6478-7d39-48ea-fb41-9df60826f655"
      },
      "source": [
        "batch_size=250\n",
        "\n",
        "train_loader = torch.utils.data.DataLoader(\n",
        "    datasets.MNIST(\"../data\",train=True,download=True,\n",
        "                   transform=transforms.Compose(\n",
        "                       [transforms.ToTensor(),\n",
        "                        transforms.Normalize((0.1307), (0.3081)),])\n",
        "                   ), batch_size=batch_size, shuffle=True)\n",
        "test_loader = torch.utils.data.DataLoader(\n",
        "    datasets.MNIST(\"../data\",train=False,\n",
        "                   transform=transforms.Compose(\n",
        "                       [transforms.ToTensor(),\n",
        "                        transforms.Normalize((0.1307), (0.3081)),])\n",
        "                   ), batch_size=1000, shuffle=False)\n"
      ],
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Downloading http://yann.lecun.com/exdb/mnist/train-images-idx3-ubyte.gz\n",
            "Failed to download (trying next):\n",
            "HTTP Error 503: Service Unavailable\n",
            "\n",
            "Downloading https://ossci-datasets.s3.amazonaws.com/mnist/train-images-idx3-ubyte.gz\n",
            "Downloading https://ossci-datasets.s3.amazonaws.com/mnist/train-images-idx3-ubyte.gz to ../data/MNIST/raw/train-images-idx3-ubyte.gz\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "3b0c428edd0b41fab0a8567e2b98a241",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=0.0, max=9912422.0), HTML(value='')))"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n",
            "Extracting ../data/MNIST/raw/train-images-idx3-ubyte.gz to ../data/MNIST/raw\n",
            "\n",
            "Downloading http://yann.lecun.com/exdb/mnist/train-labels-idx1-ubyte.gz\n",
            "Failed to download (trying next):\n",
            "HTTP Error 503: Service Unavailable\n",
            "\n",
            "Downloading https://ossci-datasets.s3.amazonaws.com/mnist/train-labels-idx1-ubyte.gz\n",
            "Downloading https://ossci-datasets.s3.amazonaws.com/mnist/train-labels-idx1-ubyte.gz to ../data/MNIST/raw/train-labels-idx1-ubyte.gz\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "c2b5ba655f184aa5b1d5a8bb250e08f0",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=0.0, max=28881.0), HTML(value='')))"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n",
            "Extracting ../data/MNIST/raw/train-labels-idx1-ubyte.gz to ../data/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 ../data/MNIST/raw/t10k-images-idx3-ubyte.gz\n",
            "Failed to download (trying next):\n",
            "HTTP Error 503: Service Unavailable\n",
            "\n",
            "Downloading https://ossci-datasets.s3.amazonaws.com/mnist/t10k-images-idx3-ubyte.gz\n",
            "Downloading https://ossci-datasets.s3.amazonaws.com/mnist/t10k-images-idx3-ubyte.gz to ../data/MNIST/raw/t10k-images-idx3-ubyte.gz\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "83d9a715473c4cdaa6759ab2d100352e",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=0.0, max=1648877.0), HTML(value='')))"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n",
            "Extracting ../data/MNIST/raw/t10k-images-idx3-ubyte.gz to ../data/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 ../data/MNIST/raw/t10k-labels-idx1-ubyte.gz\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "f56c46fd31e846fea05b864f7e22fa04",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=0.0, max=4542.0), HTML(value='')))"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n",
            "Extracting ../data/MNIST/raw/t10k-labels-idx1-ubyte.gz to ../data/MNIST/raw\n",
            "\n",
            "Processing...\n",
            "Done!\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/torchvision/datasets/mnist.py:502: UserWarning: The given NumPy array is not writeable, and PyTorch does not support non-writeable tensors. This means you can write to the underlying (supposedly non-writeable) NumPy array using the tensor. You may want to copy the array to protect its data or make it writeable before converting it to a tensor. This type of warning will be suppressed for the rest of this program. (Triggered internally at  /pytorch/torch/csrc/utils/tensor_numpy.cpp:143.)\n",
            "  return torch.from_numpy(parsed.astype(m[2], copy=False)).view(*s)\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "cZVcagU6pkgd"
      },
      "source": [
        "### We define a simple CNN, almost same as [Tensorflow Privacy example](https://github.com/tensorflow/privacy/blob/master/tutorials/Classification_Privacy.ipynb) and [Opacus example](https://github.com/pytorch/opacus/blob/master/examples/mnist.py), but we use **tanh** activation instead of **relu** for better performance (see https://arxiv.org/abs/2007.14191)."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-tLWsY8IqD9M"
      },
      "source": [
        "class SampleConvNet(nn.Module):\n",
        "    def __init__(self):\n",
        "        super().__init__()\n",
        "        self.conv1 = nn.Conv2d(1, 16, 8, 2, padding=3)\n",
        "        self.conv2 = nn.Conv2d(16, 32, 4, 2)\n",
        "        self.fc1 = nn.Linear(32 * 4 * 4, 32)\n",
        "        self.fc2 = nn.Linear(32, 10)\n",
        "\n",
        "    def forward(self, x):\n",
        "        # x of shape [B, 1, 28, 28]\n",
        "        x = F.tanh(self.conv1(x))  # -> [B, 16, 14, 14]\n",
        "        x = F.max_pool2d(x, 2, 1)  # -> [B, 16, 13, 13]\n",
        "        x = F.tanh(self.conv2(x))  # -> [B, 32, 5, 5]\n",
        "        x = F.max_pool2d(x, 2, 1)  # -> [B, 32, 4, 4]\n",
        "        x = x.view(-1, 32 * 4 * 4)  # -> [B, 512]\n",
        "        x = F.tanh(self.fc1(x))  # -> [B, 32]\n",
        "        x = self.fc2(x)  # -> [B, 10]\n",
        "        return x\n",
        "\n",
        "    def name(self):\n",
        "        return \"SampleConvNet\"\n",
        "\n",
        "device = torch.device('cuda')\n",
        "cnn_DP=SampleConvNet().to(device)"
      ],
      "execution_count": 9,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "w-Ninjk_rGiO"
      },
      "source": [
        "### Now that we have the data preprocessed and the model defined, we move on to set the training process for the optimizer. We define a class to include the hyperparameters."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "LD6pV0ZvYwiD"
      },
      "source": [
        "class Args:\n",
        "    epochs=15 # number of epochs\n",
        "    n_runs=1 # allow multiple independent run so you can take average of accuracy\n",
        "    lr=0.25 # learning rate\n",
        "    sigma=1.3 # noise multiplier in the DP optimizer, necessary to set to non-zero for DP\n",
        "    max_per_sample_grad_norm=1.5  # clipping norm to guarantee the sensitivity of gradients; necessary to set to non-infinity for DP\n",
        "    delta=1e-5 # delta in (epsilon, delta)-DP, need to be smaller than 1/sample size\n",
        "    \n",
        "    disable_dp=False  \n",
        "    # if True, then this is training with no clipping (hence not DP); \n",
        "    # if False, and if sigma not zero, then DP; else, this is non-DP training with clipping\n",
        "    \n",
        "args = Args()"
      ],
      "execution_count": 10,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "njUmKFM8sRv1"
      },
      "source": [
        "### Next, we define the training function and attach any regular optimizer (SGD, Adagrad, Adam) to the **privacy engine** (which automatically turns optimizer into DP-optimizer, and calculates your privacy risk on the fly)."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "4vC19q5FonPQ",
        "outputId": "508fa598-8836-4ed3-d017-f80a7d194936"
      },
      "source": [
        "def epoch(loader, model, opt=None):\n",
        "    total_loss, total_err = 0.,0.\n",
        "    if opt:\n",
        "      model.train()\n",
        "    else:\n",
        "      model.eval()\n",
        "    for X,y in loader:\n",
        "        X,y = X.to(device), y.to(device)\n",
        "        yp = model(X)\n",
        "        loss = nn.CrossEntropyLoss()(yp,y)\n",
        "        if opt:\n",
        "            opt.zero_grad()\n",
        "            loss.backward()\n",
        "            opt.step()\n",
        "        \n",
        "        total_err += (yp.max(dim=1)[1] != y).sum().item()\n",
        "        total_loss += loss.item() * X.shape[0]\n",
        "    return total_err / len(loader.dataset), total_loss / len(loader.dataset)\n",
        "\n",
        "opt_DP = optim.SGD(cnn_DP.parameters(), lr=args.lr)\n",
        "\n",
        "# only this privacy engine is from Opacus\n",
        "privacy_engine = PrivacyEngine(\n",
        "    cnn_DP,\n",
        "    batch_size=batch_size,\n",
        "    sample_size=len(train_loader.dataset),\n",
        "    alphas=[1 + x / 10.0 for x in range(1, 100)] + list(range(12, 64)),\n",
        "    noise_multiplier=args.sigma,\n",
        "    max_grad_norm=args.max_per_sample_grad_norm,\n",
        ")\n",
        "privacy_engine.attach(opt_DP)\n",
        "    \n",
        "print(' '*10,'Train Accuracy , Train Cross-Entropy Loss, Test Accuracy , Test Cross-Entropy Loss, ')\n",
        "for Epoch in range(args.epochs):\n",
        "    train_err_DP, train_loss_DP = epoch(train_loader, cnn_DP, opt_DP)\n",
        "    test_err_DP, test_loss_DP = epoch(test_loader, cnn_DP)\n",
        "    print('Epoch',Epoch+1,'--> DP:',*(\"{:.6f}\".format(i) for i in (1-train_err_DP, train_loss_DP, 1-test_err_DP, test_loss_DP)), sep=\"\\t\")\n",
        "    epsilon, best_alpha = privacy_engine.get_privacy_spent()\n",
        "    print(f\" | (ε = {epsilon:.2f}, δ = {privacy_engine.target_delta}) for α = {best_alpha}\")"
      ],
      "execution_count": 11,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/opacus/privacy_engine.py:518: UserWarning: The sample rate will be defined from ``batch_size`` and ``sample_size``.The returned privacy budget will be incorrect.\n",
            "  \"The sample rate will be defined from ``batch_size`` and ``sample_size``.\"\n",
            "/usr/local/lib/python3.7/dist-packages/opacus/privacy_engine.py:195: UserWarning: Secure RNG turned off. This is perfectly fine for experimentation as it allows for much faster training performance, but remember to turn it on and retrain one last time before production with ``secure_rng`` turned on.\n",
            "  \"Secure RNG turned off. This is perfectly fine for experimentation as it allows \"\n",
            "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py:795: UserWarning: Using a non-full backward hook when the forward contains multiple autograd Nodes is deprecated and will be removed in future versions. This hook will be missing some grad_input. Please use register_full_backward_hook to get the documented behavior.\n",
            "  warnings.warn(\"Using a non-full backward hook when the forward contains multiple autograd Nodes \"\n",
            "/usr/local/lib/python3.7/dist-packages/torch/nn/functional.py:1698: UserWarning: nn.functional.tanh is deprecated. Use torch.tanh instead.\n",
            "  warnings.warn(\"nn.functional.tanh is deprecated. Use torch.tanh instead.\")\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "           Train Accuracy , Train Cross-Entropy Loss, Test Accuracy , Test Cross-Entropy Loss, \n",
            "Epoch\t1\t--> DP:\t0.748200\t0.914032\t0.896900\t0.352260\n",
            " | (ε = 0.85, δ = 1e-06) for α = 18.0\n",
            "Epoch\t2\t--> DP:\t0.912100\t0.293044\t0.930300\t0.229040\n",
            " | (ε = 0.89, δ = 1e-06) for α = 18.0\n",
            "Epoch\t3\t--> DP:\t0.932617\t0.222917\t0.944700\t0.188370\n",
            " | (ε = 0.93, δ = 1e-06) for α = 18.0\n",
            "Epoch\t4\t--> DP:\t0.942767\t0.189695\t0.948000\t0.167144\n",
            " | (ε = 0.96, δ = 1e-06) for α = 18.0\n",
            "Epoch\t5\t--> DP:\t0.948000\t0.174212\t0.952300\t0.160660\n",
            " | (ε = 1.00, δ = 1e-06) for α = 18.0\n",
            "Epoch\t6\t--> DP:\t0.951833\t0.165832\t0.954100\t0.154810\n",
            " | (ε = 1.04, δ = 1e-06) for α = 18.0\n",
            "Epoch\t7\t--> DP:\t0.953317\t0.160666\t0.956400\t0.148192\n",
            " | (ε = 1.08, δ = 1e-06) for α = 18.0\n",
            "Epoch\t8\t--> DP:\t0.955383\t0.152665\t0.957000\t0.146496\n",
            " | (ε = 1.11, δ = 1e-06) for α = 17.0\n",
            "Epoch\t9\t--> DP:\t0.957617\t0.145775\t0.960600\t0.136925\n",
            " | (ε = 1.14, δ = 1e-06) for α = 17.0\n",
            "Epoch\t10\t--> DP:\t0.958733\t0.142418\t0.961700\t0.134736\n",
            " | (ε = 1.17, δ = 1e-06) for α = 17.0\n",
            "Epoch\t11\t--> DP:\t0.960133\t0.139846\t0.960800\t0.129903\n",
            " | (ε = 1.20, δ = 1e-06) for α = 17.0\n",
            "Epoch\t12\t--> DP:\t0.961800\t0.134974\t0.962700\t0.126481\n",
            " | (ε = 1.23, δ = 1e-06) for α = 17.0\n",
            "Epoch\t13\t--> DP:\t0.961783\t0.135208\t0.963700\t0.124951\n",
            " | (ε = 1.26, δ = 1e-06) for α = 17.0\n",
            "Epoch\t14\t--> DP:\t0.962050\t0.133748\t0.963200\t0.126549\n",
            " | (ε = 1.29, δ = 1e-06) for α = 17.0\n",
            "Epoch\t15\t--> DP:\t0.961733\t0.132108\t0.964000\t0.124382\n",
            " | (ε = 1.32, δ = 1e-06) for α = 17.0\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "yZAhRBhCvusH"
      },
      "source": [
        "### There exist many privacy accountants that give you an upper bound on epsilon. The most popular ones are [Moments Accountant](https://arxiv.org/abs/1607.00133) (MA) and Gaussian DP (GDP; promoted by this paper).\n",
        "\n",
        "### There are many variants of MA. E.g. in the above, Opacus uses a version of MA that gives epsilon = 1.32. In below, Tensorflow Privacy uses another MA that gives epsilon = 0.94 and our GDP accountant gives epsilon = 0.82. Hence GDP is the tightest privacy guarantee."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "67xQUH1Nt4r2",
        "outputId": "4c4fe551-9006-43f6-baed-e865c5d674a0"
      },
      "source": [
        "!pip install tensorflow-privacy\n",
        "from gdp_accountant import compute_epsP, compute_epsilon # which uses tensorflow-privacy package\n",
        "print('eps_MA=%.6g' %(compute_epsilon(15,1.3,60000,250,1e-5)))\n",
        "print('eps_GDP=%.6g' %(compute_epsP(15,1.3,60000,250,1e-5)))"
      ],
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Collecting tensorflow-privacy\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/5e/53/31a388f82202a155f248f75cc0f45bd0b85a0ef020a2472e600dc19d38d6/tensorflow_privacy-0.5.2-py3-none-any.whl (192kB)\n",
            "\r\u001b[K     |█▊                              | 10kB 16.4MB/s eta 0:00:01\r\u001b[K     |███▍                            | 20kB 22.6MB/s eta 0:00:01\r\u001b[K     |█████                           | 30kB 27.5MB/s eta 0:00:01\r\u001b[K     |██████▉                         | 40kB 21.8MB/s eta 0:00:01\r\u001b[K     |████████▌                       | 51kB 16.1MB/s eta 0:00:01\r\u001b[K     |██████████▏                     | 61kB 18.1MB/s eta 0:00:01\r\u001b[K     |████████████                    | 71kB 15.7MB/s eta 0:00:01\r\u001b[K     |█████████████▋                  | 81kB 14.1MB/s eta 0:00:01\r\u001b[K     |███████████████▎                | 92kB 14.7MB/s eta 0:00:01\r\u001b[K     |█████████████████               | 102kB 13.2MB/s eta 0:00:01\r\u001b[K     |██████████████████▊             | 112kB 13.2MB/s eta 0:00:01\r\u001b[K     |████████████████████▍           | 122kB 13.2MB/s eta 0:00:01\r\u001b[K     |██████████████████████▏         | 133kB 13.2MB/s eta 0:00:01\r\u001b[K     |███████████████████████▉        | 143kB 13.2MB/s eta 0:00:01\r\u001b[K     |█████████████████████████▌      | 153kB 13.2MB/s eta 0:00:01\r\u001b[K     |███████████████████████████▏    | 163kB 13.2MB/s eta 0:00:01\r\u001b[K     |█████████████████████████████   | 174kB 13.2MB/s eta 0:00:01\r\u001b[K     |██████████████████████████████▋ | 184kB 13.2MB/s eta 0:00:01\r\u001b[K     |████████████████████████████████| 194kB 13.2MB/s \n",
            "\u001b[?25hRequirement already satisfied: mpmath in /usr/local/lib/python3.7/dist-packages (from tensorflow-privacy) (1.2.1)\n",
            "Requirement already satisfied: dm-tree~=0.1.1 in /usr/local/lib/python3.7/dist-packages (from tensorflow-privacy) (0.1.6)\n",
            "Requirement already satisfied: tensorflow-estimator>=2.3.0 in /usr/local/lib/python3.7/dist-packages (from tensorflow-privacy) (2.4.0)\n",
            "Requirement already satisfied: scipy>=0.17 in /usr/local/lib/python3.7/dist-packages (from tensorflow-privacy) (1.4.1)\n",
            "Requirement already satisfied: six>=1.12.0 in /usr/local/lib/python3.7/dist-packages (from dm-tree~=0.1.1->tensorflow-privacy) (1.15.0)\n",
            "Requirement already satisfied: numpy>=1.13.3 in /usr/local/lib/python3.7/dist-packages (from scipy>=0.17->tensorflow-privacy) (1.19.5)\n",
            "Installing collected packages: tensorflow-privacy\n",
            "Successfully installed tensorflow-privacy-0.5.2\n",
            "eps_MA=0.9422\n",
            "eps_GDP=0.823696\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "0ZishKjos1U1"
      },
      "source": [
        "## Some Tips:\n",
        "\n",
        "1. Larger epsilon and delta means model is less private\n",
        "2. Fixing other hyperparameters, more iterations/epochs less private\n",
        "3. Learning rate and clipping norm do not affect DP but affect convergence\n",
        "4. Iteration, sigma, batch size and sample size affect DP\n",
        "5. One set of training corresponds to infinitely many (epsilon,delta) pairs, usually we choose delta = 1/sample size\n",
        "6. Practionally, one tunes clipping norm first with sigma = 0, then fix clipping norm and tune sigma, under pre-specified privacy budget (e.g. epsilon < 8)\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "oHRBD4HEjLSK"
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}