Pattern-Exploiting Training for NLP!

Поделиться
HTML-код
  • Опубликовано: 19 ноя 2024

Комментарии • 13

  • @connor-shorten
    @connor-shorten  4 года назад +4

    1:25 Pattern-Verbalizer Pairs
    2:33 HuggingFace’s NLP Viewer and Next Word Prediction Visualization
    4:40 Description of the Task and Next Word Prediction Visualization
    7:10 Patterns Tested in the Ppaer (Yelp, News Classification, Yahoo Questions, MNLI)
    10:36 Avoiding Catastrophic Forgetting
    11:45 Combining PVPs
    12:48 Iterative PET
    14:05 Experimental Results
    16:10 Automatic Verbalizer Search
    18:01 Quick Comparison wtih Semi-Supervised state-of-the-art in Computer Vision

  • @athenaspark
    @athenaspark Год назад

    Thx a lot it helped me a lot to understand one of the most impacting technique for a winning solution in FB3 kaggle competition

  • @vinayakpandi5967
    @vinayakpandi5967 3 года назад +3

    Great Video Sir...it would be of great help if you could put a tutorial of how to train our own PET.

  • @ehdrbdpdua
    @ehdrbdpdua 4 года назад +1

    Best ai youtube channel ever!

  • @sofianikitaki8417
    @sofianikitaki8417 4 года назад +2

    Great presentation and demo. Could you please share the link for the demo?

    • @connor-shorten
      @connor-shorten  4 года назад +1

      Thank you! The demo is linked in the description of the video as "Next Word Prediction"

  • @Tferdz
    @Tferdz 4 года назад +1

    Good job with the demo!

  • @КириллПопов-е1я
    @КириллПопов-е1я 4 года назад +1

    Hi! Thank you for educational content! I wonder if you can give beginners some reccomendations about how to start learning AI from scratch (maybe some books or courses u took before). All the best and more subscribers!

    • @connor-shorten
      @connor-shorten  4 года назад +2

      Thank you! I recommend starting out with Andrew Ng's lectures on youtube and some keras tutorials!

  • @mahshidhosseini9503
    @mahshidhosseini9503 4 года назад +1

    This review is amazing! So, we can use this PET approach to pre-train a language model on our specific domain and by defining a verbalizer to map our labels to natural language words, we can annotate a large set of unlabeled data. Right?

    • @connor-shorten
      @connor-shorten  4 года назад

      Thank you so much! Yes! However, finding these verbalizers can be a bit of a headache depending on the task

  • @jeanetteinthisorn4955
    @jeanetteinthisorn4955 2 года назад

    NLP is an acronym that was already in use. This was not what I wanted.