As far as my knowledge goes, many processing steps combine the three detectors' observations into a single image and is then passed to an image neural network.
\nAre there other and better ways to do this? For example having the concatenation happen inside the model instead of during the data processing step?
\nI guess the best way to know is by trying. 😄
", "messages": [ { "id": 1503330, "postDate": "2021-09-05T09:10:15.117Z", "content": "As far as my knowledge goes, many processing steps combine the three detectors' observations into a single image and is then passed to an image neural network.
\nAre there other and better ways to do this? For example having the concatenation happen inside the model instead of during the data processing step?
\nI guess the best way to know is by trying. 😄
", "rawMarkdown": "As far as my knowledge goes, many processing steps combine the three detectors' observations into a single image and is then passed to an image neural network. \n\nAre there other and better ways to do this? For example having the concatenation happen inside the model instead of during the data processing step?\n\nI guess the best way to know is by trying. 😄", "votes": 8 }, { "id": 1503913, "postDate": "2021-09-05T20:58:30.647Z", "content": "I have signals directly going to my model, perform CQT, and stacked them side by side, and (B, H, W, 1) goes to CNN.
\nRGB didn't work that great for me.
If using Keras, you can use 3 CQT for each signal and use Concatenate layer to stack them horizontally or vertically.
\nIf pytorch, 3 CQT and torch.hstack or torch.vstack should do the job.
or you can join 3 signals into one signal and do CQT once.
", "rawMarkdown": "I have signals directly going to my model, perform CQT, and stacked them side by side, and (B, H, W, 1) goes to CNN.\nRGB didn't work that great for me.\n\nIf using Keras, you can use 3 CQT for each signal and use Concatenate layer to stack them horizontally or vertically.\nIf pytorch, 3 CQT and torch.hstack or torch.vstack should do the job.\n\nor you can join 3 signals into one signal and do CQT once.", "votes": 1 }, { "id": 1559962, "postDate": "2021-10-27T08:38:55.637Z", "content": "Hey All,
\nThank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.
\nThank you all,
\nRegards,
\nChris
On a somewhat related note, since a gravitational wave should cause correlated signal across the 3 detectors, I was thinking it might be an SNR boost to use a method like wavelet coherence (e.g. matlab's wcoherence, pycwt stuff) to combine signal across detector pairs. I gave this a whirl last week and did not have much luck getting useful coherence results - compared with CWT, for example - some resolution issue is biting me.
", "rawMarkdown": "On a somewhat related note, since a gravitational wave should cause correlated signal across the 3 detectors, I was thinking it might be an SNR boost to use a method like wavelet coherence (e.g. matlab's wcoherence, pycwt stuff) to combine signal across detector pairs. I gave this a whirl last week and did not have much luck getting useful coherence results - compared with CWT, for example - some resolution issue is biting me.", "replies": [ { "id": 1505011, "postDate": "2021-09-06T20:37:03.280Z", "content": "Two reasons your experiment might have ended that way:
\nI was hoping there would be a way for a NNet with attention to learn how to correct the shift between detectors waveforms, but the reality is, it isn't even just a phase shift. You can try late-stage fusing of features, but it seems that really balloons training times… and at this point, I doubt many teams are doing that.
", "rawMarkdown": "Two reasons your experiment might have ended that way:\n- the gw doesn't hit each detector location simultaneously, so correlation as a measure might not be the best decision, but who knows\n- the gw source vs detector orientation alignment will necessarily be different at each detector location\n\nI was hoping there would be a way for a NNet with attention to learn how to correct the shift between detectors waveforms, but the reality is, it isn't even just a phase shift. You can try late-stage fusing of features, but it seems that really balloons training times... and at this point, I doubt many teams are doing that.", "votes": 3 }, { "id": 1505046, "postDate": "2021-09-06T21:28:25.827Z", "content": "In principle, coherence has phase to account for time shifts, just as cross-correlation has lag…
", "rawMarkdown": " In principle, coherence has phase to account for time shifts, just as cross-correlation has lag...", "votes": 1 } ] }, { "id": 1505044, "postDate": "2021-09-06T21:26:25.163Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 1503913, "author_name": "Harsh Patel", "author_url": "", "post_date": "2021-09-05T20:58:30.647000", "content": "I have signals directly going to my model, perform CQT, and stacked them side by side, and (B, H, W, 1) goes to CNN.
\nRGB didn't work that great for me.
If using Keras, you can use 3 CQT for each signal and use Concatenate layer to stack them horizontally or vertically.
\nIf pytorch, 3 CQT and torch.hstack or torch.vstack should do the job.
or you can join 3 signals into one signal and do CQT once.
", "votes": 1, "replies": [] }, { "id": 1559962, "author_name": "ChristopherZerafa", "author_url": "", "post_date": "2021-10-27T08:38:55.637000", "content": "Hey All,
\nThank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.
\nThank you all,
\nRegards,
\nChris
On a somewhat related note, since a gravitational wave should cause correlated signal across the 3 detectors, I was thinking it might be an SNR boost to use a method like wavelet coherence (e.g. matlab's wcoherence, pycwt stuff) to combine signal across detector pairs. I gave this a whirl last week and did not have much luck getting useful coherence results - compared with CWT, for example - some resolution issue is biting me.
", "votes": 0, "replies": [ { "id": 1505011, "author_name": "عثمان", "author_url": "", "post_date": "2021-09-06T20:37:03.280000", "content": "Two reasons your experiment might have ended that way:
\nI was hoping there would be a way for a NNet with attention to learn how to correct the shift between detectors waveforms, but the reality is, it isn't even just a phase shift. You can try late-stage fusing of features, but it seems that really balloons training times… and at this point, I doubt many teams are doing that.
", "votes": 3, "replies": [] }, { "id": 1505046, "author_name": "Stuart Johnson", "author_url": "", "post_date": "2021-09-06T21:28:25.827000", "content": "In principle, coherence has phase to account for time shifts, just as cross-correlation has lag…
", "votes": 1, "replies": [] } ] }, { "id": 1505044, "author_name": "", "author_url": "", "post_date": "2021-09-06T21:26:25.163000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1503330": "As far as my knowledge goes, many processing steps combine the three detectors' observations into a single image and is then passed to an image neural network. \n\nAre there other and better ways to do this? For example having the concatenation happen inside the model instead of during the data processing step?\n\nI guess the best way to know is by trying. 😄", "1503913": "I have signals directly going to my model, perform CQT, and stacked them side by side, and (B, H, W, 1) goes to CNN.\nRGB didn't work that great for me.\n\nIf using Keras, you can use 3 CQT for each signal and use Concatenate layer to stack them horizontally or vertically.\nIf pytorch, 3 CQT and torch.hstack or torch.vstack should do the job.\n\nor you can join 3 signals into one signal and do CQT once.", "1559962": "Hey All,\n\nThank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.\n\nThank you all,\n\nRegards,\nChris", "1504904": "On a somewhat related note, since a gravitational wave should cause correlated signal across the 3 detectors, I was thinking it might be an SNR boost to use a method like wavelet coherence (e.g. matlab's wcoherence, pycwt stuff) to combine signal across detector pairs. I gave this a whirl last week and did not have much luck getting useful coherence results - compared with CWT, for example - some resolution issue is biting me.", "1505044": "" } }