{
  "id": 133234,
  "title": "[Local loss] may be a great idea for this competition",
  "url": "/competitions/bengaliai-cv19/discussion/133234",
  "author_name": "seefun",
  "post_date": "2020-03-01T15:08:10.270000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Training neural networks without global back-propagation maybe a great idea for handwritten grapheme classification.  Seems like attention in back-propagation but not attention mechanism.</p>\n\n<p><a href=\"https://arxiv.org/abs/1901.06656\">paper here</a></p>\n\n<p><a href=\"https://github.com/anokland/local-loss\">code here</a></p>",
  "messages": [
    {
      "id": 760647,
      "postDate": "2020-03-01T15:08:10.270Z",
      "content": "<p>Training neural networks without global back-propagation maybe a great idea for handwritten grapheme classification.  Seems like attention in back-propagation but not attention mechanism.</p>\n\n<p><a href=\"https://arxiv.org/abs/1901.06656\">paper here</a></p>\n\n<p><a href=\"https://github.com/anokland/local-loss\">code here</a></p>",
      "rawMarkdown": "Training neural networks without global back-propagation maybe a great idea for handwritten grapheme classification.  Seems like attention in back-propagation but not attention mechanism.\n\n[paper here](https://arxiv.org/abs/1901.06656)\n\n[code here](https://github.com/anokland/local-loss)",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "760647": "Training neural networks without global back-propagation maybe a great idea for handwritten grapheme classification.  Seems like attention in back-propagation but not attention mechanism.\n\n[paper here](https://arxiv.org/abs/1901.06656)\n\n[code here](https://github.com/anokland/local-loss)"
  }
}