Update part of word embeddings during training

3

you can keep one separate Embedding layer for functional words, and one separate Embedding layer for non-functional words. Seems like a cleaner solution no?

1 Like