{
  "id": 241266,
  "title": "(Fast) Dispersion Measure Transform -> FDMT",
  "url": "/competitions/seti-breakthrough-listen/discussion/241266",
  "author_name": "AgentAuers",
  "post_date": "2021-05-23T20:08:07.524000",
  "votes": 12,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Barak Zackay describes in this paper a fast version of a transformation called <em>Fast Dispersion Measure Transform (FDMT)</em> to search fast radio bursts (FRB):<br>\n<a href=\"https://arxiv.org/pdf/1411.5373.pdf\" target=\"_blank\">https://arxiv.org/pdf/1411.5373.pdf</a></p>\n<p>For signal processing the dispersion measure transform looks promising. Perhapes we could transform all examples of the dataset using this to create additional \"images\" and use them as an additional channel for the input of the CNNs.</p>\n<p>There are several implementations on Github:<br>\n<a href=\"https://github.com/iansbrown/FRB-FDMT-Search\" target=\"_blank\">https://github.com/iansbrown/FRB-FDMT-Search</a><br>\n<a href=\"https://github.com/ajosephy/FDMT\" target=\"_blank\">https://github.com/ajosephy/FDMT</a><br>\n<a href=\"https://github.com/caseyjlaw/rtpipe/blob/master/rtpipe/FDMT.py\" target=\"_blank\">https://github.com/caseyjlaw/rtpipe/blob/master/rtpipe/FDMT.py</a></p>\n<p>What do you think? Is this transformation useful or not? I think this transformation is not \"stronger\" than CNNs, because if it is there would be no need for this competition. :-) But I think it can add additional information for our models to use.</p>",
  "messages": [
    {
      "id": 1320165,
      "postDate": "2021-05-23T20:08:07.523Z",
      "content": "<p>Barak Zackay describes in this paper a fast version of a transformation called <em>Fast Dispersion Measure Transform (FDMT)</em> to search fast radio bursts (FRB):<br>\n<a href=\"https://arxiv.org/pdf/1411.5373.pdf\" target=\"_blank\">https://arxiv.org/pdf/1411.5373.pdf</a></p>\n<p>For signal processing the dispersion measure transform looks promising. Perhapes we could transform all examples of the dataset using this to create additional \"images\" and use them as an additional channel for the input of the CNNs.</p>\n<p>There are several implementations on Github:<br>\n<a href=\"https://github.com/iansbrown/FRB-FDMT-Search\" target=\"_blank\">https://github.com/iansbrown/FRB-FDMT-Search</a><br>\n<a href=\"https://github.com/ajosephy/FDMT\" target=\"_blank\">https://github.com/ajosephy/FDMT</a><br>\n<a href=\"https://github.com/caseyjlaw/rtpipe/blob/master/rtpipe/FDMT.py\" target=\"_blank\">https://github.com/caseyjlaw/rtpipe/blob/master/rtpipe/FDMT.py</a></p>\n<p>What do you think? Is this transformation useful or not? I think this transformation is not \"stronger\" than CNNs, because if it is there would be no need for this competition. :-) But I think it can add additional information for our models to use.</p>",
      "rawMarkdown": "Barak Zackay describes in this paper a fast version of a transformation called *Fast Dispersion Measure Transform (FDMT)* to search fast radio bursts (FRB):\n[https://arxiv.org/pdf/1411.5373.pdf](https://arxiv.org/pdf/1411.5373.pdf)\n\nFor signal processing the dispersion measure transform looks promising. Perhapes we could transform all examples of the dataset using this to create additional \"images\" and use them as an additional channel for the input of the CNNs.\n\nThere are several implementations on Github:\n[https://github.com/iansbrown/FRB-FDMT-Search](https://github.com/iansbrown/FRB-FDMT-Search)\n[https://github.com/ajosephy/FDMT](https://github.com/ajosephy/FDMT)\n[https://github.com/caseyjlaw/rtpipe/blob/master/rtpipe/FDMT.py](https://github.com/caseyjlaw/rtpipe/blob/master/rtpipe/FDMT.py)\n\nWhat do you think? Is this transformation useful or not? I think this transformation is not \"stronger\" than CNNs, because if it is there would be no need for this competition. :-) But I think it can add additional information for our models to use.",
      "votes": 12
    },
    {
      "id": 1364019,
      "postDate": "2021-06-24T14:46:26.110Z",
      "content": "<p>did you receive any performance boost?<br>\nThe data is already Fourier transformed. Do you suggest experimenting with adding this method on top of it or some how remove Fourier transform and then use this method?</p>",
      "rawMarkdown": "did you receive any performance boost?\nThe data is already Fourier transformed. Do you suggest experimenting with adding this method on top of it or some how remove Fourier transform and then use this method?"
    }
  ],
  "comments": [
    {
      "id": 1364019,
      "author_name": "Sagar",
      "author_url": "",
      "post_date": "2021-06-24T14:46:26.110000",
      "content": "<p>did you receive any performance boost?<br>\nThe data is already Fourier transformed. Do you suggest experimenting with adding this method on top of it or some how remove Fourier transform and then use this method?</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1320165": "Barak Zackay describes in this paper a fast version of a transformation called *Fast Dispersion Measure Transform (FDMT)* to search fast radio bursts (FRB):\n[https://arxiv.org/pdf/1411.5373.pdf](https://arxiv.org/pdf/1411.5373.pdf)\n\nFor signal processing the dispersion measure transform looks promising. Perhapes we could transform all examples of the dataset using this to create additional \"images\" and use them as an additional channel for the input of the CNNs.\n\nThere are several implementations on Github:\n[https://github.com/iansbrown/FRB-FDMT-Search](https://github.com/iansbrown/FRB-FDMT-Search)\n[https://github.com/ajosephy/FDMT](https://github.com/ajosephy/FDMT)\n[https://github.com/caseyjlaw/rtpipe/blob/master/rtpipe/FDMT.py](https://github.com/caseyjlaw/rtpipe/blob/master/rtpipe/FDMT.py)\n\nWhat do you think? Is this transformation useful or not? I think this transformation is not \"stronger\" than CNNs, because if it is there would be no need for this competition. :-) But I think it can add additional information for our models to use.",
    "1364019": "did you receive any performance boost?\nThe data is already Fourier transformed. Do you suggest experimenting with adding this method on top of it or some how remove Fourier transform and then use this method?"
  }
}