“for a counterexample to your point.”
Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings
找出词嵌入中的性别方向并去除性别刻板印象,同时保留有用关联
论文指出即使是用谷歌新闻训练的词嵌入也表现出严重的男女性别刻板印象,并且其广泛使用往往会放大这些偏见。作者表明性别偏见在嵌入空间中对应一个方向,性别中性词与性别定义词线性可分,据此给出修改嵌入以消除刻板印象(例如 receptionist 与 female 的关联)、同时保留 queen 与 female 这类合理关联的方法,并定义了直接与间接性别偏见的度量。众包评估和标准基准显示,该方法显著降低了性别偏见,并保留聚类相关概念和类比任务的能力。
HN 上的人把它作为词嵌入会编码并放大性别偏见、以及可以通过去除性别方向来缓解的代表性例子,同时也有人用它批评“去偏见 AI”的做法。
学习路径
读这篇之前,按顺序读书架上的 2 篇 · 参考文献来自 OpenAlex,顺序和要点由 Claude Code 排出先理解 word2vec 词向量及其类比性质,再看如何在这个向量空间里找出并去除性别偏见。
- 1
- 2
负采样与降采样版 word2vec,目标论文用的正是这类向量
读Skip-gram 的负采样训练,以及短语向量的类比
- 3Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings
目标论文:找出词嵌入的性别方向并去偏
读性别方向的几何刻画,以及硬去偏与软去偏的做法
HN 书评
推荐在前,同类按时间 · 每篇最多保留 20 条,统计用全部提及“we now know of flaws in their outputs, such as [1].”
“The way to fight AI bias is deliberately and artificially, like in Bolukbasi et al. [1]”
“the model will learn that gender difference as part of the representation of the word, pretty much baking it into all applications. [1]”
“See Bolukbasi”
“they explicitly reject that approach (see appendix A)”
看全部 8 条 →
“Linked a few below for those interested.”
“For example the work Google does on "de-biasing AI" is all about taking ML models and warping its understanding of the world to reflect ideological priorities.”
- Debiasing Word Embeddings ↗simonster · 2016-07-25 · 2 分
- Man Is to Computer Programmer as Woman Is to Homemaker? Debiasing Word Embedding ↗yamaneko · 2016-07-27 · 2 分
- Man to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings ↗aaronyy · 2017-01-09 · 1 分