{
  "id": 612782,
  "title": "Questions & approach on Semi Supervised Learning",
  "url": "/competitions/MABe-mouse-behavior-detection/discussion/612782",
  "author_name": "",
  "post_date": "2025-10-22T04:37:07.358379800Z",
  "votes": 4,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi everyone, as we know, around 90% of the dataset is unlabeled, which means we can’t afford to ignore such a large portion without significant data loss. On the other hand, only about 10% (roughly 800 video IDs) are labeled, and relying solely on that small subset leads to overfitting.</p>\n<p>The best approach that comes to mind is semi-supervised learning (I’m sure many of you thought of this too!). I’m considering using a wrapper method to generate pseudo-labels for the unlabeled data and include them in training.</p>\n<p>Would love to hear your thoughts or suggestions on this approach!</p>",
  "messages": [
    {
      "id": "3305131",
      "postDate": "10/22/2025 04:37:07",
      "content": "<p>Hi everyone, as we know, around 90% of the dataset is unlabeled, which means we can’t afford to ignore such a large portion without significant data loss. On the other hand, only about 10% (roughly 800 video IDs) are labeled, and relying solely on that small subset leads to overfitting.</p>\n<p>The best approach that comes to mind is semi-supervised learning (I’m sure many of you thought of this too!). I’m considering using a wrapper method to generate pseudo-labels for the unlabeled data and include them in training.</p>\n<p>Would love to hear your thoughts or suggestions on this approach!</p>",
      "rawMarkdown": "Hi everyone, as we know, around 90% of the dataset is unlabeled, which means we can’t afford to ignore such a large portion without significant data loss. On the other hand, only about 10% (roughly 800 video IDs) are labeled, and relying solely on that small subset leads to overfitting.\n\nThe best approach that comes to mind is semi-supervised learning (I’m sure many of you thought of this too!). I’m considering using a wrapper method to generate pseudo-labels for the unlabeled data and include them in training.\n\nWould love to hear your thoughts or suggestions on this approach!",
      "votes": null
    },
    {
      "id": "3523115",
      "postDate": "09/10/2026 16:01:42",
      "content": "<p>Imagine training a behavior classifier with only some examples labeled. <strong>What could the rest teach it?</strong> 👀</p>\n<p><a href=\"https://mlguidance.blogspot.com/2026/09/what-are-models-of-semi-supervised.html\" target=\"_blank\">https://mlguidance.blogspot.com/2026/09/what-are-models-of-semi-supervised.html</a></p>",
      "rawMarkdown": "Imagine training a behavior classifier with only some examples labeled. **What could the rest teach it?** 👀\n\nhttps://mlguidance.blogspot.com/2026/09/what-are-models-of-semi-supervised.html",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3523115,
      "author_name": "kaustubh994",
      "author_url": "",
      "post_date": "09/10/2026 16:01:42",
      "content": "<p>Imagine training a behavior classifier with only some examples labeled. <strong>What could the rest teach it?</strong> 👀</p>\n<p><a href=\"https://mlguidance.blogspot.com/2026/09/what-are-models-of-semi-supervised.html\" target=\"_blank\">https://mlguidance.blogspot.com/2026/09/what-are-models-of-semi-supervised.html</a></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3305131": "Hi everyone, as we know, around 90% of the dataset is unlabeled, which means we can’t afford to ignore such a large portion without significant data loss. On the other hand, only about 10% (roughly 800 video IDs) are labeled, and relying solely on that small subset leads to overfitting.\n\nThe best approach that comes to mind is semi-supervised learning (I’m sure many of you thought of this too!). I’m considering using a wrapper method to generate pseudo-labels for the unlabeled data and include them in training.\n\nWould love to hear your thoughts or suggestions on this approach!",
    "3523115": "Imagine training a behavior classifier with only some examples labeled. **What could the rest teach it?** 👀\n\nhttps://mlguidance.blogspot.com/2026/09/what-are-models-of-semi-supervised.html"
  },
  "source": "meta"
}