BriefGPT.xyz
Jun, 2019
使用WEAT评估的概念消除词表示偏见
Conceptor Debiasing of Word Representations Evaluated on WEAT
HTML
PDF
Saket Karve, Lyle Ungar, João Sedoc
TL;DR
通过使用概念器去偏置来后处理传统和上下文的单词嵌入，该方法可以同时消除种族和性别偏见，并且可以有效地利用偏见单词的异构列表。该方法可以减少单词嵌入所表示的种族和性别偏见，其中通过 Caliskan 等人的单词嵌入关联测试（WEAT）来衡量。
Abstract
bias
in
word embeddings
such as Word2Vec has been widely investigated, and many efforts made to remove such
bias
. We show how to use
→