{
  "id": 363841,
  "title": "Is there any expert in the field of gravitational waves or signal matching as a participant?",
  "url": "/competitions/g2net-detecting-continuous-gravitational-waves/discussion/363841",
  "author_name": "Chen Lin",
  "post_date": "2022-11-03T11:11:46.867000",
  "votes": 5,
  "comment_count": 0,
  "views": 0,
  "content": "<p>If you are also participating in the Novozymes Enzyme Stability Prediction competition, you will find that the current high LB solutions in this competition benefit from a lot of domain knowledge rather than ML methods. I guess to some extent this competition is similar to the Novozymes Enzyme Stability Prediction competition. The task is not easy for ML, and traditional methods still perform better than NN methods. If we want to achieve higher LB or be able to implement more clever ML models, we need to understand more about traditional signal matching methods than simply using convolution as an encoder. I guess it's like relying entirely on sequence to predict the thermal stability of a protein, yes we can achieve predictions on some samples with high SNR but still can't address the low SNR part.</p>\n<p>We need experts with domain knowledge to step up and provide some ideas for everyone.</p>",
  "messages": [
    {
      "id": 2015528,
      "postDate": "2022-11-03T11:11:46.867Z",
      "content": "<p>If you are also participating in the Novozymes Enzyme Stability Prediction competition, you will find that the current high LB solutions in this competition benefit from a lot of domain knowledge rather than ML methods. I guess to some extent this competition is similar to the Novozymes Enzyme Stability Prediction competition. The task is not easy for ML, and traditional methods still perform better than NN methods. If we want to achieve higher LB or be able to implement more clever ML models, we need to understand more about traditional signal matching methods than simply using convolution as an encoder. I guess it's like relying entirely on sequence to predict the thermal stability of a protein, yes we can achieve predictions on some samples with high SNR but still can't address the low SNR part.</p>\n<p>We need experts with domain knowledge to step up and provide some ideas for everyone.</p>",
      "rawMarkdown": "If you are also participating in the Novozymes Enzyme Stability Prediction competition, you will find that the current high LB solutions in this competition benefit from a lot of domain knowledge rather than ML methods. I guess to some extent this competition is similar to the Novozymes Enzyme Stability Prediction competition. The task is not easy for ML, and traditional methods still perform better than NN methods. If we want to achieve higher LB or be able to implement more clever ML models, we need to understand more about traditional signal matching methods than simply using convolution as an encoder. I guess it's like relying entirely on sequence to predict the thermal stability of a protein, yes we can achieve predictions on some samples with high SNR but still can't address the low SNR part.\n\nWe need experts with domain knowledge to step up and provide some ideas for everyone.",
      "votes": 5
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2015528": "If you are also participating in the Novozymes Enzyme Stability Prediction competition, you will find that the current high LB solutions in this competition benefit from a lot of domain knowledge rather than ML methods. I guess to some extent this competition is similar to the Novozymes Enzyme Stability Prediction competition. The task is not easy for ML, and traditional methods still perform better than NN methods. If we want to achieve higher LB or be able to implement more clever ML models, we need to understand more about traditional signal matching methods than simply using convolution as an encoder. I guess it's like relying entirely on sequence to predict the thermal stability of a protein, yes we can achieve predictions on some samples with high SNR but still can't address the low SNR part.\n\nWe need experts with domain knowledge to step up and provide some ideas for everyone."
  }
}