This website requires JavaScript.

Stochastic Methods for AUC Optimization subject to AUC-based Fairness Constraints

Yao YaoQihang LinTianbao Yang
Dec 2022
摘要
As machine learning being used increasingly in making high-stakes decisions,an arising challenge is to avoid unfair AI systems that lead to discriminatorydecisions for protected population. A direct approach for obtaining a fairpredictive model is to train the model through optimizing its predictionperformance subject to fairness constraints, which achieves Pareto efficiencywhen trading off performance against fairness. Among various fairness metrics,the ones based on the area under the ROC curve (AUC) are emerging recentlybecause they are threshold-agnostic and effective for unbalanced data. In thiswork, we formulate the training problem of a fairness-aware machine learningmodel as an AUC optimization problem subject to a class of AUC-based fairnessconstraints. This problem can be reformulated as a min-max optimization problemwith min-max constraints, which we solve by stochastic first-order methodsbased on a new Bregman divergence designed for the special structure of theproblem. We numerically demonstrate the effectiveness of our approach onreal-world data under different fairness metrics.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?
0
被引用
笔记
问答