{
  "id": 429281,
  "title": "CTC Loss on sparse labels is Unstable",
  "url": "/competitions/asl-fingerspelling/discussion/429281",
  "author_name": "Adriano Passos",
  "post_date": "2023-08-04T20:13:27.043000",
  "votes": 4,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I am struggling to use the CTC loss with sparse labels. The computational speed for that version is from 2 to 10x faster than using dense labels. However, my training always diverge. I found other people having the same issues online but couldn't find a solution. Does any1 have any trick to solve this?</p>",
  "messages": [
    {
      "id": 2374280,
      "postDate": "2023-08-04T20:13:27.043Z",
      "content": "<p>I am struggling to use the CTC loss with sparse labels. The computational speed for that version is from 2 to 10x faster than using dense labels. However, my training always diverge. I found other people having the same issues online but couldn't find a solution. Does any1 have any trick to solve this?</p>",
      "rawMarkdown": "I am struggling to use the CTC loss with sparse labels. The computational speed for that version is from 2 to 10x faster than using dense labels. However, my training always diverge. I found other people having the same issues online but couldn't find a solution. Does any1 have any trick to solve this?",
      "votes": 3
    },
    {
      "id": 2375530,
      "postDate": "2023-08-05T17:43:17.457Z",
      "content": "<p>Could you share your implementation <a href=\"https://www.kaggle.com/coldfir3\" target=\"_blank\">@coldfir3</a>?</p>",
      "rawMarkdown": "Could you share your implementation @coldfir3?",
      "replies": [
        {
          "id": 2398230,
          "postDate": "2023-08-19T14:44:15.877Z",
          "content": "<p>Sorry for the late reply. Here: <a href=\"https://www.kaggle.com/code/coldfir3/ctc-loss/edit/run/138904672\" target=\"_blank\">https://www.kaggle.com/code/coldfir3/ctc-loss/edit/run/138904672</a></p>",
          "rawMarkdown": "Sorry for the late reply. Here: https://www.kaggle.com/code/coldfir3/ctc-loss/edit/run/138904672"
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 2375530,
      "author_name": "Yassine Alouini",
      "author_url": "",
      "post_date": "2023-08-05T17:43:17.457000",
      "content": "<p>Could you share your implementation <a href=\"https://www.kaggle.com/coldfir3\" target=\"_blank\">@coldfir3</a>?</p>",
      "votes": 0,
      "replies": [
        {
          "id": 2398230,
          "author_name": "Adriano Passos",
          "author_url": "",
          "post_date": "2023-08-19T14:44:15.877000",
          "content": "<p>Sorry for the late reply. Here: <a href=\"https://www.kaggle.com/code/coldfir3/ctc-loss/edit/run/138904672\" target=\"_blank\">https://www.kaggle.com/code/coldfir3/ctc-loss/edit/run/138904672</a></p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2374280": "I am struggling to use the CTC loss with sparse labels. The computational speed for that version is from 2 to 10x faster than using dense labels. However, my training always diverge. I found other people having the same issues online but couldn't find a solution. Does any1 have any trick to solve this?",
    "2375530": "Could you share your implementation @coldfir3?"
  }
}