{
  "id": 214053,
  "title": "How to train your model when you cannot trust on the annotations?",
  "url": "/competitions/cassava-leaf-disease-classification/discussion/214053",
  "author_name": "Hanson0910",
  "post_date": "2021-01-25T06:59:10.914000",
  "votes": 3,
  "comment_count": 3,
  "views": 0,
  "content": "<p><strong>The paper \"A Survey on Deep Learning with Noisy Labels\" analyzes how to learn robust models in noise data sets from the following four parts:</strong><br>\n<strong>1.Noise Transition Matrix<br>\n2.Robust Losses<br>\n3.Sample Weighting<br>\n4.Meta Learning<br></strong><br>\nYou can learn more about it through the following links：<br>\n<a href=\"https://arxiv.org/abs/2012.03061\" target=\"_blank\">https://arxiv.org/abs/2012.03061</a></p>",
  "messages": [
    {
      "id": 1168776,
      "postDate": "2021-01-25T06:59:10.913Z",
      "content": "<p><strong>The paper \"A Survey on Deep Learning with Noisy Labels\" analyzes how to learn robust models in noise data sets from the following four parts:</strong><br>\n<strong>1.Noise Transition Matrix<br>\n2.Robust Losses<br>\n3.Sample Weighting<br>\n4.Meta Learning<br></strong><br>\nYou can learn more about it through the following links：<br>\n<a href=\"https://arxiv.org/abs/2012.03061\" target=\"_blank\">https://arxiv.org/abs/2012.03061</a></p>",
      "rawMarkdown": "**The paper \"A Survey on Deep Learning with Noisy Labels\" analyzes how to learn robust models in noise data sets from the following four parts:**\n**1.Noise Transition Matrix<br/>\n2.Robust Losses<br/>\n3.Sample Weighting<br/>\n4.Meta Learning<br/>**\nYou can learn more about it through the following links：\nhttps://arxiv.org/abs/2012.03061",
      "votes": 3
    },
    {
      "id": 1171272,
      "postDate": "2021-01-26T18:34:49.353Z",
      "content": "<p>I am not getting this, if we make our model robust to noisy labels, then how we can expect good performance from them on the private dataset, because private dataset is noisy too.</p>\n<p><a href=\"https://www.kaggle.com/hanson0910\" target=\"_blank\">@hanson0910</a> Can you please explain this?</p>",
      "rawMarkdown": "I am not getting this, if we make our model robust to noisy labels, then how we can expect good performance from them on the private dataset, because private dataset is noisy too.\n\n@hanson0910 Can you please explain this?",
      "replies": [
        {
          "id": 1172989,
          "postDate": "2021-01-27T16:30:58.123Z",
          "rawMarkdown": "",
          "isDeleted": true
        },
        {
          "id": 1173174,
          "postDate": "2021-01-27T17:56:02.080Z",
          "content": "<p>I think, both private and public lb, includes noisy labels.</p>",
          "rawMarkdown": "I think, both private and public lb, includes noisy labels.",
          "votes": 1
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 1171272,
      "author_name": "Kishan Joshi",
      "author_url": "",
      "post_date": "2021-01-26T18:34:49.353000",
      "content": "<p>I am not getting this, if we make our model robust to noisy labels, then how we can expect good performance from them on the private dataset, because private dataset is noisy too.</p>\n<p><a href=\"https://www.kaggle.com/hanson0910\" target=\"_blank\">@hanson0910</a> Can you please explain this?</p>",
      "votes": 0,
      "replies": [
        {
          "id": 1172989,
          "author_name": "",
          "author_url": "",
          "post_date": "2021-01-27T16:30:58.123000",
          "content": "",
          "votes": 0,
          "replies": []
        },
        {
          "id": 1173174,
          "author_name": "Kishan Joshi",
          "author_url": "",
          "post_date": "2021-01-27T17:56:02.080000",
          "content": "<p>I think, both private and public lb, includes noisy labels.</p>",
          "votes": 1,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1168776": "**The paper \"A Survey on Deep Learning with Noisy Labels\" analyzes how to learn robust models in noise data sets from the following four parts:**\n**1.Noise Transition Matrix<br/>\n2.Robust Losses<br/>\n3.Sample Weighting<br/>\n4.Meta Learning<br/>**\nYou can learn more about it through the following links：\nhttps://arxiv.org/abs/2012.03061",
    "1171272": "I am not getting this, if we make our model robust to noisy labels, then how we can expect good performance from them on the private dataset, because private dataset is noisy too.\n\n@hanson0910 Can you please explain this?"
  }
}