{
  "id": 211568,
  "title": "[How to] Train student model for Learning with Noisy Labels(TPU)",
  "url": "/competitions/cassava-leaf-disease-classification/discussion/211568",
  "author_name": "",
  "post_date": "2021-01-15T16:34:48.465038200Z",
  "votes": 12,
  "comment_count": 2,
  "views": 0,
  "content": "<h1>Introduction</h1>\n<p>The paper of <code>Robustness of Accuracy Metric and its Inspirations in Learning with Noisy Labels</code> is introduced in this thread.<br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3492127%2Fc5f621fb29ee7edaf0ffcdad4a82784c%2F151515.png?generation=1610727988058077&amp;alt=media\" alt=\"\"></p>\n<p><a href=\"https://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/209318#1142756\" target=\"_blank\">https://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/209318#1142756</a></p>\n<p>I implemented it according to the cassava competition by referring to the <a href=\"https://arxiv.org/pdf/2012.04193.pdf\" target=\"_blank\">paper</a> and <a href=\"https://github.com/chenpf1025/RobustnessAccuracy\" target=\"_blank\">github</a>.</p>\n<h1>Train student model</h1>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3492127%2F149a8b3d072aba97fbb6e74e9b3f43d4%2F22222.png?generation=1610728707009346&amp;alt=media\" alt=\"\"></p>\n<ul>\n<li>Using OOF (Teacher model results)</li>\n<li>5 Folds</li>\n<li>TPU or GPU</li>\n<li>Various Loss functions</li>\n</ul>\n<h1>End</h1>\n<p>Notebook is <a href=\"https://www.kaggle.com/piantic/learning-with-noisy-labels-pytorch-xla-tpu\" target=\"_blank\">here</a>.<br>\n<a href=\"https://www.kaggle.com/piantic/learning-with-noisy-labels-pytorch-xla-tpu\" target=\"_blank\">https://www.kaggle.com/piantic/learning-with-noisy-labels-pytorch-xla-tpu</a></p>\n<p>Thank you!</p>",
  "messages": [
    {
      "id": "1154451",
      "postDate": "01/15/2021 16:34:48",
      "content": "<h1>Introduction</h1>\n<p>The paper of <code>Robustness of Accuracy Metric and its Inspirations in Learning with Noisy Labels</code> is introduced in this thread.<br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3492127%2Fc5f621fb29ee7edaf0ffcdad4a82784c%2F151515.png?generation=1610727988058077&amp;alt=media\" alt=\"\"></p>\n<p><a href=\"https://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/209318#1142756\" target=\"_blank\">https://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/209318#1142756</a></p>\n<p>I implemented it according to the cassava competition by referring to the <a href=\"https://arxiv.org/pdf/2012.04193.pdf\" target=\"_blank\">paper</a> and <a href=\"https://github.com/chenpf1025/RobustnessAccuracy\" target=\"_blank\">github</a>.</p>\n<h1>Train student model</h1>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3492127%2F149a8b3d072aba97fbb6e74e9b3f43d4%2F22222.png?generation=1610728707009346&amp;alt=media\" alt=\"\"></p>\n<ul>\n<li>Using OOF (Teacher model results)</li>\n<li>5 Folds</li>\n<li>TPU or GPU</li>\n<li>Various Loss functions</li>\n</ul>\n<h1>End</h1>\n<p>Notebook is <a href=\"https://www.kaggle.com/piantic/learning-with-noisy-labels-pytorch-xla-tpu\" target=\"_blank\">here</a>.<br>\n<a href=\"https://www.kaggle.com/piantic/learning-with-noisy-labels-pytorch-xla-tpu\" target=\"_blank\">https://www.kaggle.com/piantic/learning-with-noisy-labels-pytorch-xla-tpu</a></p>\n<p>Thank you!</p>",
      "rawMarkdown": "# Introduction\nThe paper of `Robustness of Accuracy Metric and its Inspirations in Learning with Noisy Labels` is introduced in this thread.\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3492127%2Fc5f621fb29ee7edaf0ffcdad4a82784c%2F151515.png?generation=1610727988058077&alt=media)\n\nhttps://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/209318#1142756\n\nI implemented it according to the cassava competition by referring to the [paper](https://arxiv.org/pdf/2012.04193.pdf) and [github](https://github.com/chenpf1025/RobustnessAccuracy).\n\n# Train student model\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3492127%2F149a8b3d072aba97fbb6e74e9b3f43d4%2F22222.png?generation=1610728707009346&alt=media)\n- Using OOF (Teacher model results)\n- 5 Folds\n- TPU or GPU\n- Various Loss functions\n\n# End\nNotebook is [here](https://www.kaggle.com/piantic/learning-with-noisy-labels-pytorch-xla-tpu).\nhttps://www.kaggle.com/piantic/learning-with-noisy-labels-pytorch-xla-tpu\n\nThank you!",
      "votes": null
    },
    {
      "id": "1154522",
      "postDate": "01/15/2021 17:25:45",
      "content": "<p>Did this help improve your leaderboard score?</p>",
      "rawMarkdown": "Did this help improve your leaderboard score?",
      "votes": null
    },
    {
      "id": "1154534",
      "postDate": "01/15/2021 17:30:05",
      "content": "<p>There is a little bit lower than my best public lb score. It needs a little more experimentation.</p>\n<p><a href=\"https://www.kaggle.com/ayu055\" target=\"_blank\">@ayu055</a> </p>",
      "rawMarkdown": "There is a little bit lower than my best public lb score. It needs a little more experimentation.\n\n@ayu055",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1154522,
      "author_name": "ayu055",
      "author_url": "",
      "post_date": "01/15/2021 17:25:45",
      "content": "<p>Did this help improve your leaderboard score?</p>",
      "votes": null,
      "replies": [
        {
          "id": 1154534,
          "author_name": "piantic",
          "author_url": "",
          "post_date": "01/15/2021 17:30:05",
          "content": "<p>There is a little bit lower than my best public lb score. It needs a little more experimentation.</p>\n<p><a href=\"https://www.kaggle.com/ayu055\" target=\"_blank\">@ayu055</a> </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1154451": "# Introduction\nThe paper of `Robustness of Accuracy Metric and its Inspirations in Learning with Noisy Labels` is introduced in this thread.\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3492127%2Fc5f621fb29ee7edaf0ffcdad4a82784c%2F151515.png?generation=1610727988058077&alt=media)\n\nhttps://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/209318#1142756\n\nI implemented it according to the cassava competition by referring to the [paper](https://arxiv.org/pdf/2012.04193.pdf) and [github](https://github.com/chenpf1025/RobustnessAccuracy).\n\n# Train student model\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3492127%2F149a8b3d072aba97fbb6e74e9b3f43d4%2F22222.png?generation=1610728707009346&alt=media)\n- Using OOF (Teacher model results)\n- 5 Folds\n- TPU or GPU\n- Various Loss functions\n\n# End\nNotebook is [here](https://www.kaggle.com/piantic/learning-with-noisy-labels-pytorch-xla-tpu).\nhttps://www.kaggle.com/piantic/learning-with-noisy-labels-pytorch-xla-tpu\n\nThank you!",
    "1154522": "Did this help improve your leaderboard score?",
    "1154534": "There is a little bit lower than my best public lb score. It needs a little more experimentation.\n\n@ayu055"
  },
  "source": "meta"
}