{
  "id": 133810,
  "title": "Validation Accuracy higher than training accuracy",
  "url": "/competitions/flower-classification-with-tpus/discussion/133810",
  "author_name": "",
  "post_date": "2020-03-04T11:45:53.768045500Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>When training the model with augmentation methods like mixup ,cutmix we are getting validation accuracy higher than training accuracy .Though it can be intuitively explained that training data is penalized with augmentation( and in some cases dropout )while validation is not and hence having a higher accuracy, are there any other explanations for this ? Thanks in advance.</p>",
  "messages": [
    {
      "id": "763348",
      "postDate": "03/04/2020 11:45:53",
      "content": "<p>When training the model with augmentation methods like mixup ,cutmix we are getting validation accuracy higher than training accuracy .Though it can be intuitively explained that training data is penalized with augmentation( and in some cases dropout )while validation is not and hence having a higher accuracy, are there any other explanations for this ? Thanks in advance.</p>",
      "rawMarkdown": "When training the model with augmentation methods like mixup ,cutmix we are getting validation accuracy higher than training accuracy .Though it can be intuitively explained that training data is penalized with augmentation( and in some cases dropout )while validation is not and hence having a higher accuracy, are there any other explanations for this ? Thanks in advance.",
      "votes": null
    },
    {
      "id": "763449",
      "postDate": "03/04/2020 13:52:26",
      "content": "<p>Could be overfitting, try cross validation for a robust solution.</p>",
      "rawMarkdown": "Could be overfitting, try cross validation for a robust solution.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 763449,
      "author_name": "hassanamin",
      "author_url": "",
      "post_date": "03/04/2020 13:52:26",
      "content": "<p>Could be overfitting, try cross validation for a robust solution.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "763348": "When training the model with augmentation methods like mixup ,cutmix we are getting validation accuracy higher than training accuracy .Though it can be intuitively explained that training data is penalized with augmentation( and in some cases dropout )while validation is not and hence having a higher accuracy, are there any other explanations for this ? Thanks in advance.",
    "763449": "Could be overfitting, try cross validation for a robust solution."
  },
  "source": "meta"
}