{
  "id": 366364,
  "title": "Can we do transfer learning for time-series data?",
  "url": "/competitions/applications-of-deep-learning-wustlfall-2022/discussion/366364",
  "author_name": "",
  "post_date": "2022-11-15T23:17:20.583282600Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>So Module #9 has sections to do transfer learning for computer vision and natural language processing, but is it possible to do transfer learning with time-series data? </p>\n<p>My intuition is that it is not possible because the start-of-the-art models that were listed in the 9th Colab do not have a LSTM or a Temporal CNN layer (that I am aware of). It also does not seem possible to add any of these layers to the pre-trained neurons, but please point me in the right direction if I am wrong.</p>\n<p>If this is the case, then do we just estimate the ideal number of weights and layers to put in our neural network or run a bunch of model comparisons to find the best network configuration?</p>",
  "messages": [
    {
      "id": "2031180",
      "postDate": "11/15/2022 23:17:20",
      "content": "<p>So Module #9 has sections to do transfer learning for computer vision and natural language processing, but is it possible to do transfer learning with time-series data? </p>\n<p>My intuition is that it is not possible because the start-of-the-art models that were listed in the 9th Colab do not have a LSTM or a Temporal CNN layer (that I am aware of). It also does not seem possible to add any of these layers to the pre-trained neurons, but please point me in the right direction if I am wrong.</p>\n<p>If this is the case, then do we just estimate the ideal number of weights and layers to put in our neural network or run a bunch of model comparisons to find the best network configuration?</p>",
      "rawMarkdown": "So Module #9 has sections to do transfer learning for computer vision and natural language processing, but is it possible to do transfer learning with time-series data? \n\nMy intuition is that it is not possible because the start-of-the-art models that were listed in the 9th Colab do not have a LSTM or a Temporal CNN layer (that I am aware of). It also does not seem possible to add any of these layers to the pre-trained neurons, but please point me in the right direction if I am wrong.\n\nIf this is the case, then do we just estimate the ideal number of weights and layers to put in our neural network or run a bunch of model comparisons to find the best network configuration?",
      "votes": null
    },
    {
      "id": "2038990",
      "postDate": "11/21/2022 20:38:45",
      "content": "<p>I do not know of any off the shelf model that you could use for the time series part.  For NLP and the CV parts, suggest YOLO and Glove.</p>",
      "rawMarkdown": "I do not know of any off the shelf model that you could use for the time series part.  For NLP and the CV parts, suggest YOLO and Glove.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2038990,
      "author_name": "jeffheaton",
      "author_url": "",
      "post_date": "11/21/2022 20:38:45",
      "content": "<p>I do not know of any off the shelf model that you could use for the time series part.  For NLP and the CV parts, suggest YOLO and Glove.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2031180": "So Module #9 has sections to do transfer learning for computer vision and natural language processing, but is it possible to do transfer learning with time-series data? \n\nMy intuition is that it is not possible because the start-of-the-art models that were listed in the 9th Colab do not have a LSTM or a Temporal CNN layer (that I am aware of). It also does not seem possible to add any of these layers to the pre-trained neurons, but please point me in the right direction if I am wrong.\n\nIf this is the case, then do we just estimate the ideal number of weights and layers to put in our neural network or run a bunch of model comparisons to find the best network configuration?",
    "2038990": "I do not know of any off the shelf model that you could use for the time series part.  For NLP and the CV parts, suggest YOLO and Glove."
  },
  "source": "meta"
}