{
  "id": 21753,
  "title": "is the structure of the CNN very important to performance",
  "url": "/competitions/state-farm-distracted-driver-detection/discussion/21753",
  "author_name": "",
  "post_date": "2016-06-18T03:44:57.387Z",
  "votes": null,
  "comment_count": 2,
  "views": 818,
  "content": "<p>Previously I thought it only has a small improvement from my experiments of different structures (resNet, inceptionNet, LeNet, googlenet, etc..) , but then I realize today that some people has IQ of 160 and some others only have IQ of 60, using similar set of knowledge to train these two neural nets the performance on many many things can differ dramatically. And these two neural nets might looks more or less similar? So there might be lots of more thing can be done in designing new structures of the Neural net..</p>\n\n<p>What is the difference on the actual neural net of IQ 160 and IQ 60 people? More layers? or more connections? Or on the contrary, is it a simpler net so it can be trained quickly without overfitting :) ?</p>",
  "messages": [
    {
      "id": "124403",
      "postDate": "06/18/2016 03:44:57",
      "content": "<p>Previously I thought it only has a small improvement from my experiments of different structures (resNet, inceptionNet, LeNet, googlenet, etc..) , but then I realize today that some people has IQ of 160 and some others only have IQ of 60, using similar set of knowledge to train these two neural nets the performance on many many things can differ dramatically. And these two neural nets might looks more or less similar? So there might be lots of more thing can be done in designing new structures of the Neural net..</p>\n\n<p>What is the difference on the actual neural net of IQ 160 and IQ 60 people? More layers? or more connections? Or on the contrary, is it a simpler net so it can be trained quickly without overfitting :) ?</p>",
      "rawMarkdown": "Previously I thought it only has a small improvement from my experiments of different structures (resNet, inceptionNet, LeNet, googlenet, etc..) , but then I realize today that some people has IQ of 160 and some others only have IQ of 60, using similar set of knowledge to train these two neural nets the performance on many many things can differ dramatically. And these two neural nets might looks more or less similar? So there might be lots of more thing can be done in designing new structures of the Neural net..\r\n\r\nWhat is the difference on the actual neural net of IQ 160 and IQ 60 people? More layers? or more connections? Or on the contrary, is it a simpler net so it can be trained quickly without overfitting :) ?",
      "votes": null
    },
    {
      "id": "124479",
      "postDate": "06/19/2016 04:44:03",
      "content": "<p>Hi woshialex,</p>\n\n<p>Using Neural Networks, we must think about many things which affect prediction result.<br>\nSometimes, Losers' models are similar with winners' models, but they are Inferior in some way.</p>",
      "rawMarkdown": "Hi woshialex,\r\n\r\nUsing Neural Networks, we must think about many things which affect prediction result.<br>\r\nSometimes, Losers' models are similar with winners' models, but they are Inferior in some way.",
      "votes": null
    },
    {
      "id": "124500",
      "postDate": "06/19/2016 12:15:45",
      "content": "<p>@woshialex the whole ImageNet competition, which brings us advances in image recognition, is all about CNN structure :) So, yes.</p>",
      "rawMarkdown": "woshialex the whole ImageNet competition, which brings us advances in image recognition, is all about CNN structure :) So, yes.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 124479,
      "author_name": "toshik",
      "author_url": "",
      "post_date": "06/19/2016 04:44:03",
      "content": "<p>Hi woshialex,</p>\n\n<p>Using Neural Networks, we must think about many things which affect prediction result.<br>\nSometimes, Losers' models are similar with winners' models, but they are Inferior in some way.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 124500,
      "author_name": "oldufo",
      "author_url": "",
      "post_date": "06/19/2016 12:15:45",
      "content": "<p>@woshialex the whole ImageNet competition, which brings us advances in image recognition, is all about CNN structure :) So, yes.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "124403": "Previously I thought it only has a small improvement from my experiments of different structures (resNet, inceptionNet, LeNet, googlenet, etc..) , but then I realize today that some people has IQ of 160 and some others only have IQ of 60, using similar set of knowledge to train these two neural nets the performance on many many things can differ dramatically. And these two neural nets might looks more or less similar? So there might be lots of more thing can be done in designing new structures of the Neural net..\r\n\r\nWhat is the difference on the actual neural net of IQ 160 and IQ 60 people? More layers? or more connections? Or on the contrary, is it a simpler net so it can be trained quickly without overfitting :) ?",
    "124479": "Hi woshialex,\r\n\r\nUsing Neural Networks, we must think about many things which affect prediction result.<br>\r\nSometimes, Losers' models are similar with winners' models, but they are Inferior in some way.",
    "124500": "woshialex the whole ImageNet competition, which brings us advances in image recognition, is all about CNN structure :) So, yes."
  },
  "source": "meta"
}