{
  "id": 437274,
  "title": "🔥Papers with Code for this Competition",
  "url": "/competitions/child-mind-institute-detect-sleep-states/discussion/437274",
  "author_name": "Ali Abdin",
  "post_date": "2023-09-06T06:14:03.638000",
  "votes": 143,
  "comment_count": 17,
  "views": 0,
  "content": "<ol>\n<li><a href=\"https://arxiv.org/pdf/1703.04046v2.pdf\" target=\"_blank\">DeepSleepNet: a Model for Automatic Sleep Stage Scoring based on Raw Single-Channel EEG</a><ul>\n<li><strong>Code:</strong> <a href=\"https://github.com/akaraspt/deepsleepnet\" target=\"_blank\">https://github.com/akaraspt/deepsleepnet</a></li>\n<li><strong>Summary:</strong> The study introduces DeepSleepNet, a deep learning model for automatically classifying sleep stages using single-channel EEG data. Unlike traditional methods that rely on manually-engineered features, DeepSleepNet uses Convolutional Neural Networks and bidirectional LSTMs to automatically learn relevant features and transition rules between sleep stages. It performs comparably to state-of-the-art methods but offers the advantage of adaptability to different types of EEG data without requiring manual feature engineering.</li></ul></li>\n<li><a href=\"https://arxiv.org/pdf/1910.11162v1.pdf\" target=\"_blank\">U-Time: A Fully Convolutional Network for Time Series Segmentation Applied to Sleep Staging</a><ul>\n<li><strong>Code:</strong> <a href=\"https://github.com/perslev/U-Time\" target=\"_blank\">https://github.com/perslev/U-Time</a></li>\n<li><strong>Summary:</strong> The study introduces U-Time, a fully feed-forward deep learning approach designed for physiological time-series analysis, specifically for sleep data. Based on the U-Net architecture originally used for image segmentation, U-Time is easier to tune than traditional models that combine convolutional and recurrent layers. It can map sequences of arbitrary length to class labels, essentially segmenting the time-series data. When evaluated on sleep EEG datasets, U-Time performed as well as or better than current state-of-the-art models, but with greater robustness and without needing task-specific tuning.</li></ul></li>\n<li><a href=\"https://arxiv.org/pdf/1812.04079v1.pdf\" target=\"_blank\">DOSED: a deep learning approach to detect multiple sleep micro-events in EEG signal</a><ul>\n<li><strong>Code:</strong> <a href=\"https://github.com/Dreem-Organization/dosed\" target=\"_blank\">https://github.com/Dreem-Organization/dosed</a></li>\n<li><strong>Summary:</strong> The study introduces a new deep learning architecture called Dreem One Shot Event Detector (DOSED) for automating the annotation of sleep-related events in EEG time series. Unlike traditional methods that require hand-crafted features and are specific to certain events, DOSED can simultaneously predict the locations, durations, and types of various events. Inspired by object detectors like YOLO and SSD used in computer vision, it uses a convolutional neural network for feature extraction and additional modules for localization and classification. Tested on four datasets and three types of events, DOSED showed improved performance and versatility compared to current state-of-the-art methods.</li></ul></li>\n<li><a href=\"https://arxiv.org/pdf/1809.10932v3.pdf\" target=\"_blank\">SeqSleepNet: End-to-End Hierarchical Recurrent Neural Network for Sequence-to-Sequence Automatic Sleep Staging</a><ul>\n<li><strong>Code:</strong> <a href=\"https://github.com/pquochuy/SeqSleepNet\" target=\"_blank\">https://github.com/pquochuy/SeqSleepNet</a></li>\n<li><strong>Summary:</strong> The study introduces SeqSleepNet, a hierarchical recurrent neural network designed for automatic sleep staging. Unlike traditional approaches that treat the task as a simple classification problem, SeqSleepNet takes a sequence-to-sequence approach, classifying multiple epochs at once. The network uses frequency-domain filters and attention-based recurrent layers for short-term modeling, along with an additional recurrent layer for long-term modeling of sequences. Despite its complexity, the network is trained end-to-end and outperforms existing state-of-the-art models, achieving high accuracy, F1-score, and Cohen's kappa on a dataset with 200 subjects.</li></ul></li>\n<li><a href=\"https://arxiv.org/pdf/1911.03221v4.pdf\" target=\"_blank\">Dreem Open Datasets: Multi-Scored Sleep Datasets to compare Human and Automated sleep staging</a><ul>\n<li><strong>Code:</strong> <a href=\"https://github.com/Dreem-Organization/dreem-learning-open\" target=\"_blank\">https://github.com/Dreem-Organization/dreem-learning-open</a></li>\n<li><strong>Summary:</strong> The study introduces two new datasets for sleep stage classification and a framework for comparing automated methods to multiple human scorers. A new deep learning model, SimpleSleepNet, was also developed and tested. SimpleSleepNet outperformed human experts in accuracy on both datasets, suggesting that automated methods could be more effective and reliable than human evaluation for diagnosing sleep disorders. The study argues for considering automated approaches in clinical settings.</li></ul></li>\n</ol>\n<p><strong>Check out other discussions about AI trends, papers etc:</strong><br>\n<a href=\"https://www.kaggle.com/discussions/general/437852\" target=\"_blank\">Recent AI Releases &amp; Announcements</a><br>\n<a href=\"https://www.kaggle.com/discussions/general/437099\" target=\"_blank\">Trending AI Research</a><br>\n<a href=\"https://www.kaggle.com/discussions/general/437038\" target=\"_blank\">Top 5 AI Web Tools Of The Week</a></p>",
  "messages": [
    {
      "id": 2425729,
      "postDate": "2023-09-06T06:14:03.637Z",
      "content": "<ol>\n<li><a href=\"https://arxiv.org/pdf/1703.04046v2.pdf\" target=\"_blank\">DeepSleepNet: a Model for Automatic Sleep Stage Scoring based on Raw Single-Channel EEG</a><ul>\n<li><strong>Code:</strong> <a href=\"https://github.com/akaraspt/deepsleepnet\" target=\"_blank\">https://github.com/akaraspt/deepsleepnet</a></li>\n<li><strong>Summary:</strong> The study introduces DeepSleepNet, a deep learning model for automatically classifying sleep stages using single-channel EEG data. Unlike traditional methods that rely on manually-engineered features, DeepSleepNet uses Convolutional Neural Networks and bidirectional LSTMs to automatically learn relevant features and transition rules between sleep stages. It performs comparably to state-of-the-art methods but offers the advantage of adaptability to different types of EEG data without requiring manual feature engineering.</li></ul></li>\n<li><a href=\"https://arxiv.org/pdf/1910.11162v1.pdf\" target=\"_blank\">U-Time: A Fully Convolutional Network for Time Series Segmentation Applied to Sleep Staging</a><ul>\n<li><strong>Code:</strong> <a href=\"https://github.com/perslev/U-Time\" target=\"_blank\">https://github.com/perslev/U-Time</a></li>\n<li><strong>Summary:</strong> The study introduces U-Time, a fully feed-forward deep learning approach designed for physiological time-series analysis, specifically for sleep data. Based on the U-Net architecture originally used for image segmentation, U-Time is easier to tune than traditional models that combine convolutional and recurrent layers. It can map sequences of arbitrary length to class labels, essentially segmenting the time-series data. When evaluated on sleep EEG datasets, U-Time performed as well as or better than current state-of-the-art models, but with greater robustness and without needing task-specific tuning.</li></ul></li>\n<li><a href=\"https://arxiv.org/pdf/1812.04079v1.pdf\" target=\"_blank\">DOSED: a deep learning approach to detect multiple sleep micro-events in EEG signal</a><ul>\n<li><strong>Code:</strong> <a href=\"https://github.com/Dreem-Organization/dosed\" target=\"_blank\">https://github.com/Dreem-Organization/dosed</a></li>\n<li><strong>Summary:</strong> The study introduces a new deep learning architecture called Dreem One Shot Event Detector (DOSED) for automating the annotation of sleep-related events in EEG time series. Unlike traditional methods that require hand-crafted features and are specific to certain events, DOSED can simultaneously predict the locations, durations, and types of various events. Inspired by object detectors like YOLO and SSD used in computer vision, it uses a convolutional neural network for feature extraction and additional modules for localization and classification. Tested on four datasets and three types of events, DOSED showed improved performance and versatility compared to current state-of-the-art methods.</li></ul></li>\n<li><a href=\"https://arxiv.org/pdf/1809.10932v3.pdf\" target=\"_blank\">SeqSleepNet: End-to-End Hierarchical Recurrent Neural Network for Sequence-to-Sequence Automatic Sleep Staging</a><ul>\n<li><strong>Code:</strong> <a href=\"https://github.com/pquochuy/SeqSleepNet\" target=\"_blank\">https://github.com/pquochuy/SeqSleepNet</a></li>\n<li><strong>Summary:</strong> The study introduces SeqSleepNet, a hierarchical recurrent neural network designed for automatic sleep staging. Unlike traditional approaches that treat the task as a simple classification problem, SeqSleepNet takes a sequence-to-sequence approach, classifying multiple epochs at once. The network uses frequency-domain filters and attention-based recurrent layers for short-term modeling, along with an additional recurrent layer for long-term modeling of sequences. Despite its complexity, the network is trained end-to-end and outperforms existing state-of-the-art models, achieving high accuracy, F1-score, and Cohen's kappa on a dataset with 200 subjects.</li></ul></li>\n<li><a href=\"https://arxiv.org/pdf/1911.03221v4.pdf\" target=\"_blank\">Dreem Open Datasets: Multi-Scored Sleep Datasets to compare Human and Automated sleep staging</a><ul>\n<li><strong>Code:</strong> <a href=\"https://github.com/Dreem-Organization/dreem-learning-open\" target=\"_blank\">https://github.com/Dreem-Organization/dreem-learning-open</a></li>\n<li><strong>Summary:</strong> The study introduces two new datasets for sleep stage classification and a framework for comparing automated methods to multiple human scorers. A new deep learning model, SimpleSleepNet, was also developed and tested. SimpleSleepNet outperformed human experts in accuracy on both datasets, suggesting that automated methods could be more effective and reliable than human evaluation for diagnosing sleep disorders. The study argues for considering automated approaches in clinical settings.</li></ul></li>\n</ol>\n<p><strong>Check out other discussions about AI trends, papers etc:</strong><br>\n<a href=\"https://www.kaggle.com/discussions/general/437852\" target=\"_blank\">Recent AI Releases &amp; Announcements</a><br>\n<a href=\"https://www.kaggle.com/discussions/general/437099\" target=\"_blank\">Trending AI Research</a><br>\n<a href=\"https://www.kaggle.com/discussions/general/437038\" target=\"_blank\">Top 5 AI Web Tools Of The Week</a></p>",
      "rawMarkdown": "1. [DeepSleepNet: a Model for Automatic Sleep Stage Scoring based on Raw Single-Channel EEG](https://arxiv.org/pdf/1703.04046v2.pdf)\n   - **Code:** https://github.com/akaraspt/deepsleepnet\n   - **Summary:** The study introduces DeepSleepNet, a deep learning model for automatically classifying sleep stages using single-channel EEG data. Unlike traditional methods that rely on manually-engineered features, DeepSleepNet uses Convolutional Neural Networks and bidirectional LSTMs to automatically learn relevant features and transition rules between sleep stages. It performs comparably to state-of-the-art methods but offers the advantage of adaptability to different types of EEG data without requiring manual feature engineering.\n2. [U-Time: A Fully Convolutional Network for Time Series Segmentation Applied to Sleep Staging] (https://arxiv.org/pdf/1910.11162v1.pdf)\n   - **Code:** https://github.com/perslev/U-Time\n   - **Summary:** The study introduces U-Time, a fully feed-forward deep learning approach designed for physiological time-series analysis, specifically for sleep data. Based on the U-Net architecture originally used for image segmentation, U-Time is easier to tune than traditional models that combine convolutional and recurrent layers. It can map sequences of arbitrary length to class labels, essentially segmenting the time-series data. When evaluated on sleep EEG datasets, U-Time performed as well as or better than current state-of-the-art models, but with greater robustness and without needing task-specific tuning.\n3. [DOSED: a deep learning approach to detect multiple sleep micro-events in EEG signal](https://arxiv.org/pdf/1812.04079v1.pdf)\n   - **Code:** https://github.com/Dreem-Organization/dosed\n   - **Summary:** The study introduces a new deep learning architecture called Dreem One Shot Event Detector (DOSED) for automating the annotation of sleep-related events in EEG time series. Unlike traditional methods that require hand-crafted features and are specific to certain events, DOSED can simultaneously predict the locations, durations, and types of various events. Inspired by object detectors like YOLO and SSD used in computer vision, it uses a convolutional neural network for feature extraction and additional modules for localization and classification. Tested on four datasets and three types of events, DOSED showed improved performance and versatility compared to current state-of-the-art methods.\n4. [SeqSleepNet: End-to-End Hierarchical Recurrent Neural Network for Sequence-to-Sequence Automatic Sleep Staging](https://arxiv.org/pdf/1809.10932v3.pdf)\n   - **Code:** https://github.com/pquochuy/SeqSleepNet\n   - **Summary:** The study introduces SeqSleepNet, a hierarchical recurrent neural network designed for automatic sleep staging. Unlike traditional approaches that treat the task as a simple classification problem, SeqSleepNet takes a sequence-to-sequence approach, classifying multiple epochs at once. The network uses frequency-domain filters and attention-based recurrent layers for short-term modeling, along with an additional recurrent layer for long-term modeling of sequences. Despite its complexity, the network is trained end-to-end and outperforms existing state-of-the-art models, achieving high accuracy, F1-score, and Cohen's kappa on a dataset with 200 subjects.\n5. [Dreem Open Datasets: Multi-Scored Sleep Datasets to compare Human and Automated sleep staging](https://arxiv.org/pdf/1911.03221v4.pdf)\n   - **Code:** https://github.com/Dreem-Organization/dreem-learning-open\n   - **Summary:** The study introduces two new datasets for sleep stage classification and a framework for comparing automated methods to multiple human scorers. A new deep learning model, SimpleSleepNet, was also developed and tested. SimpleSleepNet outperformed human experts in accuracy on both datasets, suggesting that automated methods could be more effective and reliable than human evaluation for diagnosing sleep disorders. The study argues for considering automated approaches in clinical settings.\n\n\n**Check out other discussions about AI trends, papers etc:**\n[Recent AI Releases & Announcements](https://www.kaggle.com/discussions/general/437852)\n[Trending AI Research](https://www.kaggle.com/discussions/general/437099)\n[Top 5 AI Web Tools Of The Week](https://www.kaggle.com/discussions/general/437038)",
      "votes": 140
    },
    {
      "id": 2438541,
      "postDate": "2023-09-14T10:47:57.397Z",
      "content": "<p>Here is one more study for time series segmentation that I am trying to implement; Hourglass CNN (<a href=\"https://www.mdpi.com/2079-9292/10/15/1805\" target=\"_blank\">https://www.mdpi.com/2079-9292/10/15/1805</a>)</p>",
      "rawMarkdown": "Here is one more study for time series segmentation that I am trying to implement; Hourglass CNN (https://www.mdpi.com/2079-9292/10/15/1805)",
      "votes": 5
    },
    {
      "id": 2448416,
      "postDate": "2023-09-20T15:09:37.527Z",
      "content": "<p>I just ran across the paper <a href=\"https://www.nature.com/articles/s41598-020-79217-x\" target=\"_blank\">Sleep classification from wrist-worn accelerometer data using random forests</a> which seems very useful and they say \"These Random forest models have been made open-source to aid further research.\" Specifically:</p>\n<pre><code>Data Availability\nThe classification models developed   paper are available   access  on Zenodo29. The R30  GGIR was previously developed  the processing of accelerometer data22. We enhanced GGIR to be able to embed the sleep classification models written  Python  explained  the GGIRpackageVignette31. Specific code to use  functionality  combination with the models from  paper can be found here. The combination of the code and GGIR  allow  sleep classification and nonwear classification of raw accelerometer . This involves  extraction, pre-processing, feature extraction, and sleep or nonwear classification.\n</code></pre>",
      "rawMarkdown": "I just ran across the paper [Sleep classification from wrist-worn accelerometer data using random forests](https://www.nature.com/articles/s41598-020-79217-x) which seems very useful and they say \"These Random forest models have been made open-source to aid further research.\" Specifically:\n```\nData Availability\nThe classification models developed in this paper are available as open access data on Zenodo29. The R30 package GGIR was previously developed for the processing of accelerometer data22. We enhanced GGIR to be able to embed the sleep classification models written in Python as explained in the GGIRpackageVignette31. Specific code to use this functionality in combination with the models from this paper can be found here. The combination of the code and GGIR package allow for sleep classification and nonwear classification of raw accelerometer data. This involves data extraction, pre-processing, feature extraction, and sleep or nonwear classification.\n```",
      "votes": 6
    },
    {
      "id": 2428548,
      "postDate": "2023-09-08T02:05:19.770Z",
      "content": "<p>Thank you for sharing!<br>\n↓SOTA by datasets<br>\n<a href=\"https://paperswithcode.com/sota/automatic-sleep-stage-classification-on-sleep-1\" target=\"_blank\">https://paperswithcode.com/sota/automatic-sleep-stage-classification-on-sleep-1</a><br>\n<a href=\"https://paperswithcode.com/sota/sleep-stage-detection-on-mass-ss3\" target=\"_blank\">https://paperswithcode.com/sota/sleep-stage-detection-on-mass-ss3</a></p>",
      "rawMarkdown": "Thank you for sharing!\n↓SOTA by datasets\nhttps://paperswithcode.com/sota/automatic-sleep-stage-classification-on-sleep-1\nhttps://paperswithcode.com/sota/sleep-stage-detection-on-mass-ss3",
      "votes": 3
    },
    {
      "id": 2525744,
      "postDate": "2023-11-15T10:19:09.457Z",
      "content": "<p>Thanks for the papers! Will try to get some inspiration for good heuristics baselines and advanced architectures</p>",
      "rawMarkdown": "Thanks for the papers! Will try to get some inspiration for good heuristics baselines and advanced architectures"
    },
    {
      "id": 2509005,
      "postDate": "2023-11-02T05:55:02.950Z",
      "content": "<p>Can I ask how many timestamps per minute for the EEG dataset?</p>",
      "rawMarkdown": "Can I ask how many timestamps per minute for the EEG dataset?"
    },
    {
      "id": 2502571,
      "postDate": "2023-10-28T09:28:39.217Z",
      "content": "<p>Hi thanks for the info, Is it allowed to use these open source? From what I saw no internet connection for this competition.</p>",
      "rawMarkdown": "Hi thanks for the info, Is it allowed to use these open source? From what I saw no internet connection for this competition."
    },
    {
      "id": 2430221,
      "postDate": "2023-09-09T06:30:04.473Z",
      "content": "<p>Thank you for sharing</p>",
      "rawMarkdown": "Thank you for sharing",
      "votes": 1
    },
    {
      "id": 2428849,
      "postDate": "2023-09-08T07:41:51.850Z",
      "content": "<p>Thanks for sharing!</p>",
      "rawMarkdown": "Thanks for sharing!",
      "votes": 1
    },
    {
      "id": 2428534,
      "postDate": "2023-09-08T01:26:22.407Z",
      "content": "<p>Thanks for sharing!</p>",
      "rawMarkdown": "Thanks for sharing!",
      "votes": 1
    },
    {
      "id": 2427409,
      "postDate": "2023-09-07T07:48:43.490Z",
      "content": "<p>Thanks for your sharing</p>",
      "rawMarkdown": "Thanks for your sharing",
      "votes": 2
    },
    {
      "id": 2502304,
      "postDate": "2023-10-28T05:17:03.990Z",
      "content": "<p>Thanks for sharing this.</p>",
      "rawMarkdown": "Thanks for sharing this."
    },
    {
      "id": 2498713,
      "postDate": "2023-10-25T13:22:00.243Z",
      "content": "<p>Thanks for sharing!</p>",
      "rawMarkdown": "Thanks for sharing!"
    },
    {
      "id": 2464845,
      "postDate": "2023-10-02T14:51:42.457Z",
      "content": "<p>Thank you so much for sharing!</p>",
      "rawMarkdown": "Thank you so much for sharing!"
    },
    {
      "id": 2446044,
      "postDate": "2023-09-19T08:05:31.473Z",
      "content": "<p>Thank you for sharing!</p>",
      "rawMarkdown": "Thank you for sharing!"
    },
    {
      "id": 2446030,
      "postDate": "2023-09-19T07:52:57.723Z",
      "content": "<p>Thanks for sharing!</p>",
      "rawMarkdown": "Thanks for sharing!"
    },
    {
      "id": 2444507,
      "postDate": "2023-09-18T10:01:16.950Z",
      "content": "<p>Thank you for sharing</p>",
      "rawMarkdown": "Thank you for sharing"
    },
    {
      "id": 2434431,
      "postDate": "2023-09-12T09:45:06.310Z",
      "content": "<p>Thanks for sharing!</p>",
      "rawMarkdown": "Thanks for sharing!",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 2438541,
      "author_name": "LewsTherin",
      "author_url": "",
      "post_date": "2023-09-14T10:47:57.397000",
      "content": "<p>Here is one more study for time series segmentation that I am trying to implement; Hourglass CNN (<a href=\"https://www.mdpi.com/2079-9292/10/15/1805\" target=\"_blank\">https://www.mdpi.com/2079-9292/10/15/1805</a>)</p>",
      "votes": 5,
      "replies": []
    },
    {
      "id": 2448416,
      "author_name": "Daniel Dewey",
      "author_url": "",
      "post_date": "2023-09-20T15:09:37.527000",
      "content": "<p>I just ran across the paper <a href=\"https://www.nature.com/articles/s41598-020-79217-x\" target=\"_blank\">Sleep classification from wrist-worn accelerometer data using random forests</a> which seems very useful and they say \"These Random forest models have been made open-source to aid further research.\" Specifically:</p>\n<pre><code>Data Availability\nThe classification models developed   paper are available   access  on Zenodo29. The R30  GGIR was previously developed  the processing of accelerometer data22. We enhanced GGIR to be able to embed the sleep classification models written  Python  explained  the GGIRpackageVignette31. Specific code to use  functionality  combination with the models from  paper can be found here. The combination of the code and GGIR  allow  sleep classification and nonwear classification of raw accelerometer . This involves  extraction, pre-processing, feature extraction, and sleep or nonwear classification.\n</code></pre>",
      "votes": 6,
      "replies": []
    },
    {
      "id": 2428548,
      "author_name": "kueno",
      "author_url": "",
      "post_date": "2023-09-08T02:05:19.770000",
      "content": "<p>Thank you for sharing!<br>\n↓SOTA by datasets<br>\n<a href=\"https://paperswithcode.com/sota/automatic-sleep-stage-classification-on-sleep-1\" target=\"_blank\">https://paperswithcode.com/sota/automatic-sleep-stage-classification-on-sleep-1</a><br>\n<a href=\"https://paperswithcode.com/sota/sleep-stage-detection-on-mass-ss3\" target=\"_blank\">https://paperswithcode.com/sota/sleep-stage-detection-on-mass-ss3</a></p>",
      "votes": 3,
      "replies": []
    },
    {
      "id": 2525744,
      "author_name": "NicolasPllr1",
      "author_url": "",
      "post_date": "2023-11-15T10:19:09.457000",
      "content": "<p>Thanks for the papers! Will try to get some inspiration for good heuristics baselines and advanced architectures</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 2509005,
      "author_name": "Mingzheng Shi",
      "author_url": "",
      "post_date": "2023-11-02T05:55:02.950000",
      "content": "<p>Can I ask how many timestamps per minute for the EEG dataset?</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 2502571,
      "author_name": "Ruby Hartono",
      "author_url": "",
      "post_date": "2023-10-28T09:28:39.217000",
      "content": "<p>Hi thanks for the info, Is it allowed to use these open source? From what I saw no internet connection for this competition.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 2430221,
      "author_name": "Miguel Granados",
      "author_url": "",
      "post_date": "2023-09-09T06:30:04.473000",
      "content": "<p>Thank you for sharing</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 2428849,
      "author_name": "cogitoo",
      "author_url": "",
      "post_date": "2023-09-08T07:41:51.850000",
      "content": "<p>Thanks for sharing!</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 2428534,
      "author_name": "Mahimai Raja J",
      "author_url": "",
      "post_date": "2023-09-08T01:26:22.407000",
      "content": "<p>Thanks for sharing!</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 2427409,
      "author_name": "LeeOyeji",
      "author_url": "",
      "post_date": "2023-09-07T07:48:43.490000",
      "content": "<p>Thanks for your sharing</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 2502304,
      "author_name": "Deepak Dhull",
      "author_url": "",
      "post_date": "2023-10-28T05:17:03.990000",
      "content": "<p>Thanks for sharing this.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 2498713,
      "author_name": "Kazu",
      "author_url": "",
      "post_date": "2023-10-25T13:22:00.243000",
      "content": "<p>Thanks for sharing!</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 2464845,
      "author_name": "Álvaro Chapresto",
      "author_url": "",
      "post_date": "2023-10-02T14:51:42.457000",
      "content": "<p>Thank you so much for sharing!</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 2446044,
      "author_name": "",
      "author_url": "",
      "post_date": "2023-09-19T08:05:31.473000",
      "content": "<p>Thank you for sharing!</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 2446030,
      "author_name": "Christopher Behrends",
      "author_url": "",
      "post_date": "2023-09-19T07:52:57.723000",
      "content": "<p>Thanks for sharing!</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 2444507,
      "author_name": "Rureka",
      "author_url": "",
      "post_date": "2023-09-18T10:01:16.950000",
      "content": "<p>Thank you for sharing</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 2434431,
      "author_name": "",
      "author_url": "",
      "post_date": "2023-09-12T09:45:06.310000",
      "content": "<p>Thanks for sharing!</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2425729": "1. [DeepSleepNet: a Model for Automatic Sleep Stage Scoring based on Raw Single-Channel EEG](https://arxiv.org/pdf/1703.04046v2.pdf)\n   - **Code:** https://github.com/akaraspt/deepsleepnet\n   - **Summary:** The study introduces DeepSleepNet, a deep learning model for automatically classifying sleep stages using single-channel EEG data. Unlike traditional methods that rely on manually-engineered features, DeepSleepNet uses Convolutional Neural Networks and bidirectional LSTMs to automatically learn relevant features and transition rules between sleep stages. It performs comparably to state-of-the-art methods but offers the advantage of adaptability to different types of EEG data without requiring manual feature engineering.\n2. [U-Time: A Fully Convolutional Network for Time Series Segmentation Applied to Sleep Staging] (https://arxiv.org/pdf/1910.11162v1.pdf)\n   - **Code:** https://github.com/perslev/U-Time\n   - **Summary:** The study introduces U-Time, a fully feed-forward deep learning approach designed for physiological time-series analysis, specifically for sleep data. Based on the U-Net architecture originally used for image segmentation, U-Time is easier to tune than traditional models that combine convolutional and recurrent layers. It can map sequences of arbitrary length to class labels, essentially segmenting the time-series data. When evaluated on sleep EEG datasets, U-Time performed as well as or better than current state-of-the-art models, but with greater robustness and without needing task-specific tuning.\n3. [DOSED: a deep learning approach to detect multiple sleep micro-events in EEG signal](https://arxiv.org/pdf/1812.04079v1.pdf)\n   - **Code:** https://github.com/Dreem-Organization/dosed\n   - **Summary:** The study introduces a new deep learning architecture called Dreem One Shot Event Detector (DOSED) for automating the annotation of sleep-related events in EEG time series. Unlike traditional methods that require hand-crafted features and are specific to certain events, DOSED can simultaneously predict the locations, durations, and types of various events. Inspired by object detectors like YOLO and SSD used in computer vision, it uses a convolutional neural network for feature extraction and additional modules for localization and classification. Tested on four datasets and three types of events, DOSED showed improved performance and versatility compared to current state-of-the-art methods.\n4. [SeqSleepNet: End-to-End Hierarchical Recurrent Neural Network for Sequence-to-Sequence Automatic Sleep Staging](https://arxiv.org/pdf/1809.10932v3.pdf)\n   - **Code:** https://github.com/pquochuy/SeqSleepNet\n   - **Summary:** The study introduces SeqSleepNet, a hierarchical recurrent neural network designed for automatic sleep staging. Unlike traditional approaches that treat the task as a simple classification problem, SeqSleepNet takes a sequence-to-sequence approach, classifying multiple epochs at once. The network uses frequency-domain filters and attention-based recurrent layers for short-term modeling, along with an additional recurrent layer for long-term modeling of sequences. Despite its complexity, the network is trained end-to-end and outperforms existing state-of-the-art models, achieving high accuracy, F1-score, and Cohen's kappa on a dataset with 200 subjects.\n5. [Dreem Open Datasets: Multi-Scored Sleep Datasets to compare Human and Automated sleep staging](https://arxiv.org/pdf/1911.03221v4.pdf)\n   - **Code:** https://github.com/Dreem-Organization/dreem-learning-open\n   - **Summary:** The study introduces two new datasets for sleep stage classification and a framework for comparing automated methods to multiple human scorers. A new deep learning model, SimpleSleepNet, was also developed and tested. SimpleSleepNet outperformed human experts in accuracy on both datasets, suggesting that automated methods could be more effective and reliable than human evaluation for diagnosing sleep disorders. The study argues for considering automated approaches in clinical settings.\n\n\n**Check out other discussions about AI trends, papers etc:**\n[Recent AI Releases & Announcements](https://www.kaggle.com/discussions/general/437852)\n[Trending AI Research](https://www.kaggle.com/discussions/general/437099)\n[Top 5 AI Web Tools Of The Week](https://www.kaggle.com/discussions/general/437038)",
    "2438541": "Here is one more study for time series segmentation that I am trying to implement; Hourglass CNN (https://www.mdpi.com/2079-9292/10/15/1805)",
    "2448416": "I just ran across the paper [Sleep classification from wrist-worn accelerometer data using random forests](https://www.nature.com/articles/s41598-020-79217-x) which seems very useful and they say \"These Random forest models have been made open-source to aid further research.\" Specifically:\n```\nData Availability\nThe classification models developed in this paper are available as open access data on Zenodo29. The R30 package GGIR was previously developed for the processing of accelerometer data22. We enhanced GGIR to be able to embed the sleep classification models written in Python as explained in the GGIRpackageVignette31. Specific code to use this functionality in combination with the models from this paper can be found here. The combination of the code and GGIR package allow for sleep classification and nonwear classification of raw accelerometer data. This involves data extraction, pre-processing, feature extraction, and sleep or nonwear classification.\n```",
    "2428548": "Thank you for sharing!\n↓SOTA by datasets\nhttps://paperswithcode.com/sota/automatic-sleep-stage-classification-on-sleep-1\nhttps://paperswithcode.com/sota/sleep-stage-detection-on-mass-ss3",
    "2525744": "Thanks for the papers! Will try to get some inspiration for good heuristics baselines and advanced architectures",
    "2509005": "Can I ask how many timestamps per minute for the EEG dataset?",
    "2502571": "Hi thanks for the info, Is it allowed to use these open source? From what I saw no internet connection for this competition.",
    "2430221": "Thank you for sharing",
    "2428849": "Thanks for sharing!",
    "2428534": "Thanks for sharing!",
    "2427409": "Thanks for your sharing",
    "2502304": "Thanks for sharing this.",
    "2498713": "Thanks for sharing!",
    "2464845": "Thank you so much for sharing!",
    "2446044": "Thank you for sharing!",
    "2446030": "Thanks for sharing!",
    "2444507": "Thank you for sharing",
    "2434431": "Thanks for sharing!"
  }
}