Debiasing representations by removing unwanted variation due to protected attributes
Debiasing representations by removing unwanted variation due to protected attributes
复制标题
通过消除由于受保护的属性而导致的不需要的变化来消除表示偏差
DOI：
--
复制
发表时间：
2018
期刊：
影响因子：
--
通讯作者：
Alexander Vargo
中科院分区：
文献类型：
--
作者：
Amanda Bower;Laura Niss;Yuekai Sun;Alexander Vargo
我们提出了一种基于回归的方法来消除隐含的偏见表示。在观察到受保护属性的任务上，该方法在统计上比已知方法更有效。此外，我们表明，这种方法导致去偏表示，满足一阶近似的条件奇偶校验。最后，我们证明了所提出的方法的有效性，减少种族偏见的累犯风险分数。
We propose a regression-based approach to removing implicit biases in representations. On tasks where the protected attribute is observed, the method is statistically more efficient than known approaches. Further, we show that this approach leads to debiased representations that satisfy a first order approximation of conditional parity. Finally, we demonstrate the efficacy of the proposed approach by reducing racial bias in recidivism risk scores.