data_mining:neural_network:word_embeddings

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revisionPrevious revision
data_mining:neural_network:word_embeddings [2018/06/09 16:37] – [Debiasing word embeddings] phreazerdata_mining:neural_network:word_embeddings [2018/06/09 16:40] (current) – [Debiasing word embeddings] phreazer
Line 121: Line 121:
  
   - Identify bias direction (e.g. gender)   - Identify bias direction (e.g. gender)
-    - $e_{he} - e_{she}$, average them +    * $e_{he} - e_{she}$, average them 
-  -  +  -  Neutralize: For every word that is not definitial (legitimate gender component), project 
 +  - Equalize pairs: Only difference should be gender (e.g. grandfather vs. grandmother); equidistant
  
  
  • data_mining/neural_network/word_embeddings.1528562225.txt.gz
  • Last modified: 2018/06/09 16:37
  • by phreazer