Debiasing word embeddings

[tuananhhedspibk]

In the Equalization part: we ensure that the distance from "grandmother" or "grandfather" to "neutralized words" like (babysitter or doctor) are the same. We do that by moving "grandmother" and "grandfather" to a pair of points that are equidistant from the axis in the middle.

Ref: https://www.coursera.org/learn/nlp-sequence-models/lecture/zHASj/debiasing-word-embeddings