{
  "id": 97898,
  "title": "Google Activates 'Dropout' Patent(US9406017B2) ",
  "url": "/competitions/aptos2019-blindness-detection/discussion/97898",
  "author_name": "",
  "post_date": "2019-06-29T14:36:16.401318600Z",
  "votes": 1,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Dropout scheme is commonly used these days in almost all Deep Learning models. The beauty of this method ensures reduction of <strong>Overfitting</strong>, <strong>Low Compute</strong> utilization through <strong>Regularization</strong>. <br>\nDropout scheme was invented by the great <strong>Dr.Geoffrey Hinton</strong> and his team of data scientists from University of Toronto and acquired by Google in 2016. <br>\nWe all enjoy the DL, ML, AI and many other goodies given at no cost by Google and various communities..</p>\n\n<ul>\n<li>What it means to a common data scientists and enthusiasts like most of us?</li>\n<li>Does that mean, we cannot use Dropouts in our models?</li>\n<li>Or we can use Dropouts for academic purpose but not with commercial interests?</li>\n</ul>\n\n<p>Please share your thoughts.\n<a href=\"https://patents.google.com/patent/US9406017B2/en\">https://patents.google.com/patent/US9406017B2/en</a></p>",
  "messages": [
    {
      "id": "564515",
      "postDate": "06/29/2019 14:36:16",
      "content": "<p>Dropout scheme is commonly used these days in almost all Deep Learning models. The beauty of this method ensures reduction of <strong>Overfitting</strong>, <strong>Low Compute</strong> utilization through <strong>Regularization</strong>. <br>\nDropout scheme was invented by the great <strong>Dr.Geoffrey Hinton</strong> and his team of data scientists from University of Toronto and acquired by Google in 2016. <br>\nWe all enjoy the DL, ML, AI and many other goodies given at no cost by Google and various communities..</p>\n\n<ul>\n<li>What it means to a common data scientists and enthusiasts like most of us?</li>\n<li>Does that mean, we cannot use Dropouts in our models?</li>\n<li>Or we can use Dropouts for academic purpose but not with commercial interests?</li>\n</ul>\n\n<p>Please share your thoughts.\n<a href=\"https://patents.google.com/patent/US9406017B2/en\">https://patents.google.com/patent/US9406017B2/en</a></p>",
      "rawMarkdown": "Dropout scheme is commonly used these days in almost all Deep Learning models. The beauty of this method ensures reduction of **Overfitting**, **Low Compute** utilization through **Regularization**.  \nDropout scheme was invented by the great **Dr.Geoffrey Hinton** and his team of data scientists from University of Toronto and acquired by Google in 2016.  \nWe all enjoy the DL, ML, AI and many other goodies given at no cost by Google and various communities..\n\n- What it means to a common data scientists and enthusiasts like most of us?\n- Does that mean, we cannot use Dropouts in our models?\n- Or we can use Dropouts for academic purpose but not with commercial interests?\n\nPlease share your thoughts.\nhttps://patents.google.com/patent/US9406017B2/en",
      "votes": null
    },
    {
      "id": "564593",
      "postDate": "06/29/2019 17:10:33",
      "content": "<p>Isn't the fundamental concept behind dropouts similar to that of random pruning used in Random forests? This cannot be patented</p>",
      "rawMarkdown": "Isn't the fundamental concept behind dropouts similar to that of random pruning used in Random forests? This cannot be patented",
      "votes": null
    },
    {
      "id": "564674",
      "postDate": "06/29/2019 19:32:12",
      "content": "<p>From what I've heard, this is has no effect on basically everybody. Theoretically google <em>could</em> attempt to enforce the patent but they wont(they have many other patents on ml stuff). \nThe purpose of this is to prevent patent trolls who would patent it and actually enforce the patent, then, a lot of companies(like google) would lose money by having to license it etc.</p>",
      "rawMarkdown": "From what I've heard, this is has no effect on basically everybody. Theoretically google *could* attempt to enforce the patent but they wont(they have many other patents on ml stuff). \nThe purpose of this is to prevent patent trolls who would patent it and actually enforce the patent, then, a lot of companies(like google) would lose money by having to license it etc.",
      "votes": null
    },
    {
      "id": "564961",
      "postDate": "06/30/2019 08:45:12",
      "content": "<p>It's the same thing !</p>",
      "rawMarkdown": "It's the same thing !",
      "votes": null
    },
    {
      "id": "565755",
      "postDate": "07/01/2019 10:37:43",
      "content": "<p>Patenting this kind of basic stuff is a bit worrying, since it could affect such a huge number of people/companies using deep learning for production. I guess for Kaggle type of use it should have no effect.</p>\n\n<p>It is nice of Google to promise not to sue anyone with these patents, but you are just hanging on their word. I would prefer if a more public organization such as UofT and Hinton's group would hold them, but I guess that is business for you. And if you can patent it, and someone is willing to sell it, what can you do...</p>",
      "rawMarkdown": "Patenting this kind of basic stuff is a bit worrying, since it could affect such a huge number of people/companies using deep learning for production. I guess for Kaggle type of use it should have no effect.\n\nIt is nice of Google to promise not to sue anyone with these patents, but you are just hanging on their word. I would prefer if a more public organization such as UofT and Hinton's group would hold them, but I guess that is business for you. And if you can patent it, and someone is willing to sell it, what can you do...",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 564593,
      "author_name": "sriharipramod",
      "author_url": "",
      "post_date": "06/29/2019 17:10:33",
      "content": "<p>Isn't the fundamental concept behind dropouts similar to that of random pruning used in Random forests? This cannot be patented</p>",
      "votes": null,
      "replies": [
        {
          "id": 564961,
          "author_name": "harshthaker",
          "author_url": "",
          "post_date": "06/30/2019 08:45:12",
          "content": "<p>It's the same thing !</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 564674,
      "author_name": "sidhanthholalkere",
      "author_url": "",
      "post_date": "06/29/2019 19:32:12",
      "content": "<p>From what I've heard, this is has no effect on basically everybody. Theoretically google <em>could</em> attempt to enforce the patent but they wont(they have many other patents on ml stuff). \nThe purpose of this is to prevent patent trolls who would patent it and actually enforce the patent, then, a lot of companies(like google) would lose money by having to license it etc.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 565755,
      "author_name": "donkeys",
      "author_url": "",
      "post_date": "07/01/2019 10:37:43",
      "content": "<p>Patenting this kind of basic stuff is a bit worrying, since it could affect such a huge number of people/companies using deep learning for production. I guess for Kaggle type of use it should have no effect.</p>\n\n<p>It is nice of Google to promise not to sue anyone with these patents, but you are just hanging on their word. I would prefer if a more public organization such as UofT and Hinton's group would hold them, but I guess that is business for you. And if you can patent it, and someone is willing to sell it, what can you do...</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "564515": "Dropout scheme is commonly used these days in almost all Deep Learning models. The beauty of this method ensures reduction of **Overfitting**, **Low Compute** utilization through **Regularization**.  \nDropout scheme was invented by the great **Dr.Geoffrey Hinton** and his team of data scientists from University of Toronto and acquired by Google in 2016.  \nWe all enjoy the DL, ML, AI and many other goodies given at no cost by Google and various communities..\n\n- What it means to a common data scientists and enthusiasts like most of us?\n- Does that mean, we cannot use Dropouts in our models?\n- Or we can use Dropouts for academic purpose but not with commercial interests?\n\nPlease share your thoughts.\nhttps://patents.google.com/patent/US9406017B2/en",
    "564593": "Isn't the fundamental concept behind dropouts similar to that of random pruning used in Random forests? This cannot be patented",
    "564674": "From what I've heard, this is has no effect on basically everybody. Theoretically google *could* attempt to enforce the patent but they wont(they have many other patents on ml stuff). \nThe purpose of this is to prevent patent trolls who would patent it and actually enforce the patent, then, a lot of companies(like google) would lose money by having to license it etc.",
    "564961": "It's the same thing !",
    "565755": "Patenting this kind of basic stuff is a bit worrying, since it could affect such a huge number of people/companies using deep learning for production. I guess for Kaggle type of use it should have no effect.\n\nIt is nice of Google to promise not to sue anyone with these patents, but you are just hanging on their word. I would prefer if a more public organization such as UofT and Hinton's group would hold them, but I guess that is business for you. And if you can patent it, and someone is willing to sell it, what can you do..."
  },
  "source": "meta"
}