{
  "id": 39336,
  "title": "Human log loss for image classification (deep learning vs. human perception) ",
  "url": "/competitions/intel-mobileodt-cervical-cancer-screening/discussion/39336",
  "author_name": "Piotr Migdał",
  "post_date": "2017-09-12T10:02:29.527000",
  "votes": 3,
  "comment_count": 0,
  "views": 0,
  "content": "<p>We all know that it was hard to distinguish different cervical opening types. It turned out that for artificial neural networks the task was not much easier.</p>\n\n<p>I share one more lesson learnt from this competition - how to estimate human performance (so to set a reasonable benchmark for our neural networks):</p>\n\n<p><a href=\"https://blog.deepsense.ai/human-log-loss-for-image-classification/\">https://blog.deepsense.ai/human-log-loss-for-image-classification/</a> </p>\n\n<p>It uses conditional entropy, to translate discrete guesses into a log loss estimate. I am curious what do you think about this approach! :)</p>\n\n<p><img src=\"https://blog.deepsense.ai/wp-content/uploads/2017/09/plot4.png\" alt=\"enter image description here\" title=\"\"></p>",
  "messages": [
    {
      "id": 220490,
      "postDate": "2017-09-12T10:02:29.527Z",
      "content": "<p>We all know that it was hard to distinguish different cervical opening types. It turned out that for artificial neural networks the task was not much easier.</p>\n\n<p>I share one more lesson learnt from this competition - how to estimate human performance (so to set a reasonable benchmark for our neural networks):</p>\n\n<p><a href=\"https://blog.deepsense.ai/human-log-loss-for-image-classification/\">https://blog.deepsense.ai/human-log-loss-for-image-classification/</a> </p>\n\n<p>It uses conditional entropy, to translate discrete guesses into a log loss estimate. I am curious what do you think about this approach! :)</p>\n\n<p><img src=\"https://blog.deepsense.ai/wp-content/uploads/2017/09/plot4.png\" alt=\"enter image description here\" title=\"\"></p>",
      "rawMarkdown": "We all know that it was hard to distinguish different cervical opening types. It turned out that for artificial neural networks the task was not much easier.\n\nI share one more lesson learnt from this competition - how to estimate human performance (so to set a reasonable benchmark for our neural networks):\n\nhttps://blog.deepsense.ai/human-log-loss-for-image-classification/ \n\nIt uses conditional entropy, to translate discrete guesses into a log loss estimate. I am curious what do you think about this approach! :)\n\n![enter image description here][1]\n\n  [1]: https://blog.deepsense.ai/wp-content/uploads/2017/09/plot4.png",
      "votes": 3
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "220490": "We all know that it was hard to distinguish different cervical opening types. It turned out that for artificial neural networks the task was not much easier.\n\nI share one more lesson learnt from this competition - how to estimate human performance (so to set a reasonable benchmark for our neural networks):\n\nhttps://blog.deepsense.ai/human-log-loss-for-image-classification/ \n\nIt uses conditional entropy, to translate discrete guesses into a log loss estimate. I am curious what do you think about this approach! :)\n\n![enter image description here][1]\n\n  [1]: https://blog.deepsense.ai/wp-content/uploads/2017/09/plot4.png"
  }
}