Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

...

Expand

It represents an intuitively slightly ridiculous null hypothesis that often works surprisingly well as a feature.

Example:

 

 

 

How do we proceed during the age of deep learning, where, for prediction, we don't need to (aren't supposed to) worry about features anymore?

Expand
  1. BERT vs hand features, controversy paper
  2. Word embeddings
    (BERT - word pieces!)
  3. Language modeling = the bridge?