{
  "id": 492197,
  "title": "Attention is what we need?",
  "url": "/competitions/hms-harmful-brain-activity-classification/discussion/492197",
  "author_name": "",
  "post_date": "2024-04-09T00:34:05.927334400Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Just saw the first few top teams said that they used transformers as one of their models… I wonder how many of you used transformers? Also wonder why there were no public notebooks with transformer architectures being used that have comparable scores with the CNNs.</p>",
  "messages": [
    {
      "id": "2742519",
      "postDate": "04/09/2024 00:34:05",
      "content": "<p>Just saw the first few top teams said that they used transformers as one of their models… I wonder how many of you used transformers? Also wonder why there were no public notebooks with transformer architectures being used that have comparable scores with the CNNs.</p>",
      "rawMarkdown": "Just saw the first few top teams said that they used transformers as one of their models... I wonder how many of you used transformers? Also wonder why there were no public notebooks with transformer architectures being used that have comparable scores with the CNNs.",
      "votes": null
    },
    {
      "id": "2742739",
      "postDate": "04/09/2024 03:50:30",
      "content": "<p>Half of our models were transformers.</p>",
      "rawMarkdown": "Half of our models were transformers.",
      "votes": null
    },
    {
      "id": "2742772",
      "postDate": "04/09/2024 04:24:34",
      "content": "<p>I got a large boost (+0.01 or +0.02) switching all image models from EfficientNet (CNN model) to VIT (Transformer model).</p>",
      "rawMarkdown": "I got a large boost (+0.01 or +0.02) switching all image models from EfficientNet (CNN model) to VIT (Transformer model).",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2742739,
      "author_name": "gunesevitan",
      "author_url": "",
      "post_date": "04/09/2024 03:50:30",
      "content": "<p>Half of our models were transformers.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2742772,
      "author_name": "cdeotte",
      "author_url": "",
      "post_date": "04/09/2024 04:24:34",
      "content": "<p>I got a large boost (+0.01 or +0.02) switching all image models from EfficientNet (CNN model) to VIT (Transformer model).</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2742519": "Just saw the first few top teams said that they used transformers as one of their models... I wonder how many of you used transformers? Also wonder why there were no public notebooks with transformer architectures being used that have comparable scores with the CNNs.",
    "2742739": "Half of our models were transformers.",
    "2742772": "I got a large boost (+0.01 or +0.02) switching all image models from EfficientNet (CNN model) to VIT (Transformer model)."
  },
  "source": "meta"
}