{
  "id": 437553,
  "title": "Google just released new time series algorithm based on attention mechanism",
  "url": "/competitions/child-mind-institute-detect-sleep-states/discussion/437553",
  "author_name": "Chirag Chauhan",
  "post_date": "2023-09-07T08:08:47.532000",
  "votes": 13,
  "comment_count": 0,
  "views": 0,
  "content": "<p>So Google just released a new time series algorithm called <strong>\"TSMixer\"</strong> based on <strong>attention mechanism</strong>. Google claims that the model leverages <strong>linear model characteristics</strong> and performs well on <strong>long-term forecasting benchmarks</strong>.</p>\n<p>Google stated this in their official blog </p>\n<blockquote>\n  <p>TSMixer is the first multivariate model that performs as well as state-of-the-art univariate models on long-term forecasting benchmarks, where we show that cross-variate information is less beneficial</p>\n</blockquote>\n<p>The below image shows the difference between the <strong>Transformer and TSMixer architecture</strong>:</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5324499%2Fdbcc17536142b8163776ab1b3bb3cfff%2FScreenshot%202023-09-07%20133613.png?generation=1694074019967788&amp;alt=media\" alt=\"\"></p>\n<p><strong>Link to the official blog</strong>: <a href=\"https://blog.research.google/2023/09/tsmixer-all-mlp-architecture-for-time.html?m=1\" target=\"_blank\">https://blog.research.google/2023/09/tsmixer-all-mlp-architecture-for-time.html?m=1</a></p>",
  "messages": [
    {
      "id": 2427434,
      "postDate": "2023-09-07T08:08:47.533Z",
      "content": "<p>So Google just released a new time series algorithm called <strong>\"TSMixer\"</strong> based on <strong>attention mechanism</strong>. Google claims that the model leverages <strong>linear model characteristics</strong> and performs well on <strong>long-term forecasting benchmarks</strong>.</p>\n<p>Google stated this in their official blog </p>\n<blockquote>\n  <p>TSMixer is the first multivariate model that performs as well as state-of-the-art univariate models on long-term forecasting benchmarks, where we show that cross-variate information is less beneficial</p>\n</blockquote>\n<p>The below image shows the difference between the <strong>Transformer and TSMixer architecture</strong>:</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5324499%2Fdbcc17536142b8163776ab1b3bb3cfff%2FScreenshot%202023-09-07%20133613.png?generation=1694074019967788&amp;alt=media\" alt=\"\"></p>\n<p><strong>Link to the official blog</strong>: <a href=\"https://blog.research.google/2023/09/tsmixer-all-mlp-architecture-for-time.html?m=1\" target=\"_blank\">https://blog.research.google/2023/09/tsmixer-all-mlp-architecture-for-time.html?m=1</a></p>",
      "rawMarkdown": "So Google just released a new time series algorithm called **\"TSMixer\"** based on **attention mechanism**. Google claims that the model leverages **linear model characteristics** and performs well on **long-term forecasting benchmarks**.\n\nGoogle stated this in their official blog \n>TSMixer is the first multivariate model that performs as well as state-of-the-art univariate models on long-term forecasting benchmarks, where we show that cross-variate information is less beneficial\n\nThe below image shows the difference between the **Transformer and TSMixer architecture**:\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5324499%2Fdbcc17536142b8163776ab1b3bb3cfff%2FScreenshot%202023-09-07%20133613.png?generation=1694074019967788&alt=media)\n\n**Link to the official blog**: https://blog.research.google/2023/09/tsmixer-all-mlp-architecture-for-time.html?m=1",
      "votes": 13
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2427434": "So Google just released a new time series algorithm called **\"TSMixer\"** based on **attention mechanism**. Google claims that the model leverages **linear model characteristics** and performs well on **long-term forecasting benchmarks**.\n\nGoogle stated this in their official blog \n>TSMixer is the first multivariate model that performs as well as state-of-the-art univariate models on long-term forecasting benchmarks, where we show that cross-variate information is less beneficial\n\nThe below image shows the difference between the **Transformer and TSMixer architecture**:\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5324499%2Fdbcc17536142b8163776ab1b3bb3cfff%2FScreenshot%202023-09-07%20133613.png?generation=1694074019967788&alt=media)\n\n**Link to the official blog**: https://blog.research.google/2023/09/tsmixer-all-mlp-architecture-for-time.html?m=1"
  }
}