{
  "id": 68509,
  "title": "Interpretability of models",
  "url": "/competitions/airbus-ship-detection/discussion/68509",
  "author_name": "Liana Napalkova",
  "post_date": "2018-10-13T16:47:56.402000",
  "votes": 0,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hello,</p>\n\n<p>I wonder if someone has applied any interpretability technique in this competition? For example:</p>\n\n<p><a href=\"https://distill.pub/2018/building-blocks/\">https://distill.pub/2018/building-blocks/</a></p>\n\n<p>Basically my concern is if we are supposed to analyse the algorithmic reasoning? Different recent studies demonstrate that a model can pass all tests successfully, but it does not really mean that it learned what it was supposed to learn. </p>\n\n<p>One of such examples is a deep learning model that was supposed to learn the differences between dogs and wolves, but finally it learned to distinguish between snow and non-snow ground.</p>\n\n<p>Thanks.</p>\n\n<p>Liana</p>",
  "messages": [
    {
      "id": 403449,
      "postDate": "2018-10-13T16:47:56.403Z",
      "content": "<p>Hello,</p>\n\n<p>I wonder if someone has applied any interpretability technique in this competition? For example:</p>\n\n<p><a href=\"https://distill.pub/2018/building-blocks/\">https://distill.pub/2018/building-blocks/</a></p>\n\n<p>Basically my concern is if we are supposed to analyse the algorithmic reasoning? Different recent studies demonstrate that a model can pass all tests successfully, but it does not really mean that it learned what it was supposed to learn. </p>\n\n<p>One of such examples is a deep learning model that was supposed to learn the differences between dogs and wolves, but finally it learned to distinguish between snow and non-snow ground.</p>\n\n<p>Thanks.</p>\n\n<p>Liana</p>",
      "rawMarkdown": "Hello,\n\nI wonder if someone has applied any interpretability technique in this competition? For example:\n\n[https://distill.pub/2018/building-blocks/][1]\n\nBasically my concern is if we are supposed to analyse the algorithmic reasoning? Different recent studies demonstrate that a model can pass all tests successfully, but it does not really mean that it learned what it was supposed to learn. \n\nOne of such examples is a deep learning model that was supposed to learn the differences between dogs and wolves, but finally it learned to distinguish between snow and non-snow ground.\n\nThanks.\n\nLiana\n\n  [1]: https://distill.pub/2018/building-blocks/"
    },
    {
      "id": 577756,
      "postDate": "2019-07-17T03:34:45.153Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 577756,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-07-17T03:34:45.153000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "403449": "Hello,\n\nI wonder if someone has applied any interpretability technique in this competition? For example:\n\n[https://distill.pub/2018/building-blocks/][1]\n\nBasically my concern is if we are supposed to analyse the algorithmic reasoning? Different recent studies demonstrate that a model can pass all tests successfully, but it does not really mean that it learned what it was supposed to learn. \n\nOne of such examples is a deep learning model that was supposed to learn the differences between dogs and wolves, but finally it learned to distinguish between snow and non-snow ground.\n\nThanks.\n\nLiana\n\n  [1]: https://distill.pub/2018/building-blocks/",
    "577756": ""
  }
}