{
  "id": 47478,
  "title": "Deep Residual Learning Benchmark (95.8% on speech commands dataset)",
  "url": "/competitions/tensorflow-speech-recognition-challenge/discussion/47478",
  "author_name": "",
  "post_date": "2018-01-14T22:29:21.672386300Z",
  "votes": null,
  "comment_count": 6,
  "views": 0,
  "content": "<p>Hey, I know this is a pretty late, but has anyone tried applying architecture from <a href=\"https://arxiv.org/pdf/1710.10361.pdf\">our paper</a>? We established a 95.8% accuracy on the speech commands dataset, which I understand is slightly different... Our <a href=\"http://github.com/castorini/honk\">repository</a> has implementations of both residual and traditional models, if anyone is interested.</p>",
  "messages": [
    {
      "id": "268564",
      "postDate": "01/14/2018 22:29:21",
      "content": "<p>Hey, I know this is a pretty late, but has anyone tried applying architecture from <a href=\"https://arxiv.org/pdf/1710.10361.pdf\">our paper</a>? We established a 95.8% accuracy on the speech commands dataset, which I understand is slightly different... Our <a href=\"http://github.com/castorini/honk\">repository</a> has implementations of both residual and traditional models, if anyone is interested.</p>",
      "rawMarkdown": "Hey, I know this is a pretty late, but has anyone tried applying architecture from [our paper][1]? We established a 95.8% accuracy on the speech commands dataset, which I understand is slightly different... Our [repository][2] has implementations of both residual and traditional models, if anyone is interested.\n\n  [1]: https://arxiv.org/pdf/1710.10361.pdf\n  [2]: http://github.com/castorini/honk",
      "votes": null
    },
    {
      "id": "268572",
      "postDate": "01/14/2018 22:57:16",
      "content": "<p>I am fitting my resnet now... will report a public leaderboard score here in 5 hours.</p>\n\n<p>UPDATE: I was able to get .87 with a bag of 5 resnet-18 my validation accuracy is 0.964.</p>",
      "rawMarkdown": "I am fitting my resnet now... will report a public leaderboard score here in 5 hours.\n\nUPDATE: I was able to get .87 with a bag of 5 resnet-18 my validation accuracy is 0.964.",
      "votes": null
    },
    {
      "id": "268600",
      "postDate": "01/15/2018 00:43:20",
      "content": "<p>I tried the resnet which structure is similar with your paper. In my case, it achieved about more than 0.95 in train and validation, but only 0.82 in public LB.  </p>",
      "rawMarkdown": "I tried the resnet which structure is similar with your paper. In my case, it achieved about more than 0.95 in train and validation, but only 0.82 in public LB.",
      "votes": null
    },
    {
      "id": "268607",
      "postDate": "01/15/2018 01:00:35",
      "content": "<p>Also tried the model at the very beginning of this competition, I think the best I got was only 0.83</p>",
      "rawMarkdown": "Also tried the model at the very beginning of this competition, I think the best I got was only 0.83",
      "votes": null
    },
    {
      "id": "268690",
      "postDate": "01/15/2018 08:11:21",
      "content": "<p>Your ResNet-15 code was my starting point (thanks!) and achieved 95.6% local validation and 86% on the public Leaderboard.</p>",
      "rawMarkdown": "Your ResNet-15 code was my starting point (thanks!) and achieved 95.6% local validation and 86% on the public Leaderboard.",
      "votes": null
    },
    {
      "id": "268746",
      "postDate": "01/15/2018 13:21:21",
      "content": "<p>All our resnets scored 86% and outperformed by VGG which can reach 87% LB. </p>",
      "rawMarkdown": "All our resnets scored 86% and outperformed by VGG which can reach 87% LB.",
      "votes": null
    },
    {
      "id": "269070",
      "postDate": "01/16/2018 06:13:28",
      "content": "<p>This was <a href=\"https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/43624\">linked early on </a>in the competition by Heng CherKeng so it got some viewing.</p>",
      "rawMarkdown": "This was [linked early on ][1]in the competition by Heng CherKeng so it got some viewing.\n\n\n  [1]: https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/43624",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 268572,
      "author_name": "ryanzhang",
      "author_url": "",
      "post_date": "01/14/2018 22:57:16",
      "content": "<p>I am fitting my resnet now... will report a public leaderboard score here in 5 hours.</p>\n\n<p>UPDATE: I was able to get .87 with a bag of 5 resnet-18 my validation accuracy is 0.964.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 268600,
      "author_name": "yatzhash",
      "author_url": "",
      "post_date": "01/15/2018 00:43:20",
      "content": "<p>I tried the resnet which structure is similar with your paper. In my case, it achieved about more than 0.95 in train and validation, but only 0.82 in public LB.  </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 268607,
      "author_name": "xiaozhouwang",
      "author_url": "",
      "post_date": "01/15/2018 01:00:35",
      "content": "<p>Also tried the model at the very beginning of this competition, I think the best I got was only 0.83</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 268690,
      "author_name": "timothyman",
      "author_url": "",
      "post_date": "01/15/2018 08:11:21",
      "content": "<p>Your ResNet-15 code was my starting point (thanks!) and achieved 95.6% local validation and 86% on the public Leaderboard.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 268746,
      "author_name": "tugstugi",
      "author_url": "",
      "post_date": "01/15/2018 13:21:21",
      "content": "<p>All our resnets scored 86% and outperformed by VGG which can reach 87% LB. </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 269070,
      "author_name": "atfisnotatf",
      "author_url": "",
      "post_date": "01/16/2018 06:13:28",
      "content": "<p>This was <a href=\"https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/43624\">linked early on </a>in the competition by Heng CherKeng so it got some viewing.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "268564": "Hey, I know this is a pretty late, but has anyone tried applying architecture from [our paper][1]? We established a 95.8% accuracy on the speech commands dataset, which I understand is slightly different... Our [repository][2] has implementations of both residual and traditional models, if anyone is interested.\n\n  [1]: https://arxiv.org/pdf/1710.10361.pdf\n  [2]: http://github.com/castorini/honk",
    "268572": "I am fitting my resnet now... will report a public leaderboard score here in 5 hours.\n\nUPDATE: I was able to get .87 with a bag of 5 resnet-18 my validation accuracy is 0.964.",
    "268600": "I tried the resnet which structure is similar with your paper. In my case, it achieved about more than 0.95 in train and validation, but only 0.82 in public LB.",
    "268607": "Also tried the model at the very beginning of this competition, I think the best I got was only 0.83",
    "268690": "Your ResNet-15 code was my starting point (thanks!) and achieved 95.6% local validation and 86% on the public Leaderboard.",
    "268746": "All our resnets scored 86% and outperformed by VGG which can reach 87% LB.",
    "269070": "This was [linked early on ][1]in the competition by Heng CherKeng so it got some viewing.\n\n\n  [1]: https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/43624"
  },
  "source": "meta"
}