{
  "id": 110381,
  "title": "How will Recursion use the best solutions?",
  "url": "/competitions/recursion-cellular-image-classification/discussion/110381",
  "author_name": "",
  "post_date": "2019-09-27T08:11:25.930582100Z",
  "votes": 5,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I'm curious to know how will Recursion use the output and insights from this competition, as the mission was to \"eliminate the noise introduced by technical execution and environmental variation between experiments\".</p>",
  "messages": [
    {
      "id": "635166",
      "postDate": "09/27/2019 08:11:25",
      "content": "<p>I'm curious to know how will Recursion use the output and insights from this competition, as the mission was to \"eliminate the noise introduced by technical execution and environmental variation between experiments\".</p>",
      "rawMarkdown": "I'm curious to know how will Recursion use the output and insights from this competition, as the mission was to \"eliminate the noise introduced by technical execution and environmental variation between experiments\".",
      "votes": null
    },
    {
      "id": "635345",
      "postDate": "09/27/2019 12:06:58",
      "content": "<p>That's a good question, as the top solutions will greatly exploit the experiment setup (aka leak), with the single treatment per well and experiment.\nNevertheless, IMHO the trained models (even using pseudo labels) still know a lot about the data and could be used as a backbone for other tasks, e.g. using transfer learning...</p>",
      "rawMarkdown": "That's a good question, as the top solutions will greatly exploit the experiment setup (aka leak), with the single treatment per well and experiment.\nNevertheless, IMHO the trained models (even using pseudo labels) still know a lot about the data and could be used as a backbone for other tasks, e.g. using transfer learning...",
      "votes": null
    },
    {
      "id": "635369",
      "postDate": "09/27/2019 12:42:49",
      "content": "<p>Good observation, plus I see that there's some top models that didn't exploit the leak. </p>\n\n<p>PS The next time you use IMHO in your text, I'll downvote you.</p>",
      "rawMarkdown": "Good observation, plus I see that there's some top models that didn't exploit the leak. \n\nPS The next time you use IMHO in your text, I'll downvote you.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 635345,
      "author_name": "hmendonca",
      "author_url": "",
      "post_date": "09/27/2019 12:06:58",
      "content": "<p>That's a good question, as the top solutions will greatly exploit the experiment setup (aka leak), with the single treatment per well and experiment.\nNevertheless, IMHO the trained models (even using pseudo labels) still know a lot about the data and could be used as a backbone for other tasks, e.g. using transfer learning...</p>",
      "votes": null,
      "replies": [
        {
          "id": 635369,
          "author_name": "giuliasavorgnan",
          "author_url": "",
          "post_date": "09/27/2019 12:42:49",
          "content": "<p>Good observation, plus I see that there's some top models that didn't exploit the leak. </p>\n\n<p>PS The next time you use IMHO in your text, I'll downvote you.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "635166": "I'm curious to know how will Recursion use the output and insights from this competition, as the mission was to \"eliminate the noise introduced by technical execution and environmental variation between experiments\".",
    "635345": "That's a good question, as the top solutions will greatly exploit the experiment setup (aka leak), with the single treatment per well and experiment.\nNevertheless, IMHO the trained models (even using pseudo labels) still know a lot about the data and could be used as a backbone for other tasks, e.g. using transfer learning...",
    "635369": "Good observation, plus I see that there's some top models that didn't exploit the leak. \n\nPS The next time you use IMHO in your text, I'll downvote you."
  },
  "source": "meta"
}