Skip to content
No description, website, or topics provided.
Python
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
data
README.md
accuracies_20_5000.png
find_transform.py
nn.py
word2vec.py

README.md

Are word embeddings universal?

The latent variable space is normally non-identifiable up to an orthonormal transformation, so no. But, can we estimate that transformation given a small set of paired points?

Does it work for less explicitly paired data eg different language word embeddings? What about sentence embeddings?

Reconstruction accuracy for 20D embedding, $|V|=5000$

ToDo

  • Use better corpus eg europarl
  • Enforce orthonormality
  • Transform between languages
You can’t perform that action at this time.