{
  "id": 12360,
  "title": "Interesting Read : No more pesky learning rates",
  "url": "/competitions/datasciencebowl/discussion/12360",
  "author_name": "",
  "post_date": "2015-01-29T21:22:04.307Z",
  "votes": 1,
  "comment_count": 2,
  "views": 1870,
  "content": "<p><a href=\"http://arxiv.org/pdf/1206.1106.pdf\">http://arxiv.org/pdf/1206.1106.pdf</a></p>",
  "messages": [
    {
      "id": "63327",
      "postDate": "01/29/2015 21:22:04",
      "content": "<p><a href=\"http://arxiv.org/pdf/1206.1106.pdf\">http://arxiv.org/pdf/1206.1106.pdf</a></p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "63329",
      "postDate": "01/29/2015 21:45:51",
      "content": "<p>Yes this is a pretty nice paper. Now depending on the software you use it may be difficult to implement the method they describe because you need to compute the diagonal of the Hessian matrix.</p>\n<p>For example on torch7 it's hard to get it work with convolutionnal layers.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "63330",
      "postDate": "01/29/2015 21:54:19",
      "content": "<p>There's a follow-up and approximation method for the Hessian described here:</p>\n<p><a href=\"http://arxiv.org/pdf/1301.3764.pdf\">http://arxiv.org/pdf/1301.3764.pdf</a></p>\n<p>It's on my list of things to try. If anyone else has attempted it, please do share your experience</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 63329,
      "author_name": "irmo322",
      "author_url": "",
      "post_date": "01/29/2015 21:45:51",
      "content": "<p>Yes this is a pretty nice paper. Now depending on the software you use it may be difficult to implement the method they describe because you need to compute the diagonal of the Hessian matrix.</p>\n<p>For example on torch7 it's hard to get it work with convolutionnal layers.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 63330,
      "author_name": "ashhafez",
      "author_url": "",
      "post_date": "01/29/2015 21:54:19",
      "content": "<p>There's a follow-up and approximation method for the Hessian described here:</p>\n<p><a href=\"http://arxiv.org/pdf/1301.3764.pdf\">http://arxiv.org/pdf/1301.3764.pdf</a></p>\n<p>It's on my list of things to try. If anyone else has attempted it, please do share your experience</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "63327": "",
    "63329": "",
    "63330": ""
  },
  "source": "meta"
}