{
  "id": 363130,
  "title": "Are you using Dropout or Batch normalization for MLP??",
  "url": "/competitions/open-problems-multimodal/discussion/363130",
  "author_name": "wakaka",
  "post_date": "2022-10-31T08:30:13.754000",
  "votes": 2,
  "comment_count": 5,
  "views": 0,
  "content": "<p>Hello,<br>\nI'm wondering if I should use Dropout or batchnormalization. When I use Dropout, CV goes up but LB goes down.<br>\nWhen I use Batchnormalization, CV and LB droped. Please let me know if you are using them and why.</p>",
  "messages": [
    {
      "id": 2011973,
      "postDate": "2022-11-01T01:08:31.433Z",
      "content": "<p>Hi, my LB also decreased when I used batch normalization, and CV&amp;LB decreased. Usually batch normalization can help us meet the convergence condition more quickly, which means that we can achieve the best position in a relative short time. I think if your machine can handle a large number of epoches, batch normalization is not necessary.</p>",
      "rawMarkdown": "Hi, my LB also decreased when I used batch normalization, and CV&LB decreased. Usually batch normalization can help us meet the convergence condition more quickly, which means that we can achieve the best position in a relative short time. I think if your machine can handle a large number of epoches, batch normalization is not necessary.",
      "votes": 1,
      "replies": [
        {
          "id": 2013013,
          "postDate": "2022-11-01T15:24:47.287Z",
          "content": "<p>Thank you so much!</p>",
          "rawMarkdown": "Thank you so much!"
        }
      ]
    },
    {
      "id": 2011346,
      "postDate": "2022-10-31T14:12:55.760Z",
      "content": "<p>I use layers Dropout (0.1 and 0.4) and BatchNormalization with CV and LB improvements</p>",
      "rawMarkdown": "I use layers Dropout (0.1 and 0.4) and BatchNormalization with CV and LB improvements",
      "votes": 1,
      "replies": [
        {
          "id": 2013011,
          "postDate": "2022-11-01T15:24:16.287Z",
          "content": "<p>Thank you for your information！</p>",
          "rawMarkdown": "Thank you for your information！"
        }
      ]
    },
    {
      "id": 2010910,
      "postDate": "2022-10-31T08:30:13.753Z",
      "content": "<p>Hello,<br>\nI'm wondering if I should use Dropout or batchnormalization. When I use Dropout, CV goes up but LB goes down.<br>\nWhen I use Batchnormalization, CV and LB droped. Please let me know if you are using them and why.</p>",
      "rawMarkdown": "Hello,\nI'm wondering if I should use Dropout or batchnormalization. When I use Dropout, CV goes up but LB goes down.\nWhen I use Batchnormalization, CV and LB droped. Please let me know if you are using them and why.",
      "votes": 2
    },
    {
      "id": 2029995,
      "postDate": "2022-11-15T05:54:58.753Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 2011973,
      "author_name": "TESUZI",
      "author_url": "",
      "post_date": "2022-11-01T01:08:31.433000",
      "content": "<p>Hi, my LB also decreased when I used batch normalization, and CV&amp;LB decreased. Usually batch normalization can help us meet the convergence condition more quickly, which means that we can achieve the best position in a relative short time. I think if your machine can handle a large number of epoches, batch normalization is not necessary.</p>",
      "votes": 1,
      "replies": [
        {
          "id": 2013013,
          "author_name": "wakaka",
          "author_url": "",
          "post_date": "2022-11-01T15:24:47.287000",
          "content": "<p>Thank you so much!</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 2011346,
      "author_name": "Ricardo Colomer",
      "author_url": "",
      "post_date": "2022-10-31T14:12:55.760000",
      "content": "<p>I use layers Dropout (0.1 and 0.4) and BatchNormalization with CV and LB improvements</p>",
      "votes": 1,
      "replies": [
        {
          "id": 2013011,
          "author_name": "wakaka",
          "author_url": "",
          "post_date": "2022-11-01T15:24:16.287000",
          "content": "<p>Thank you for your information！</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 2029995,
      "author_name": "",
      "author_url": "",
      "post_date": "2022-11-15T05:54:58.753000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2011973": "Hi, my LB also decreased when I used batch normalization, and CV&LB decreased. Usually batch normalization can help us meet the convergence condition more quickly, which means that we can achieve the best position in a relative short time. I think if your machine can handle a large number of epoches, batch normalization is not necessary.",
    "2011346": "I use layers Dropout (0.1 and 0.4) and BatchNormalization with CV and LB improvements",
    "2010910": "Hello,\nI'm wondering if I should use Dropout or batchnormalization. When I use Dropout, CV goes up but LB goes down.\nWhen I use Batchnormalization, CV and LB droped. Please let me know if you are using them and why.",
    "2029995": ""
  }
}