按年被提到
20162016 · 3 次2026
机器学习与深度学习›贝叶斯方法与 AutoML·arXiv:1609.05807 · cs.LG

Inherent Trade-Offs in the Fair Determination of Risk Scores

公平确定风险评分中的内在权衡
Jon Kleinberg, Sendhil Mullainathan, Manish Raghavan · 2016

证明多种公平性条件除极特殊情形外无法同时满足

开始阅读 ↗PDFalphaXiv 讨论
被推荐 · 8 人首次 2016-11-06 · 最近 2024-06-27
8不同的人
8活跃月份
9讨论串
1,432被引(OpenAlex)

针对公众对算法分类中「对不同群体公平」含义的争论,作者形式化了处于争论核心的三种公平性条件,并证明除了高度受限的特殊情形外,没有任何方法能同时满足这三个条件。进一步地,即使只是近似同时满足,也要求数据近似落在定理所确定的某个受限特殊情形里。这表明关键的公平概念彼此不相容,并为权衡它们提供了思考框架。

HN 为什么提它

HN 上的人用它说明各种「公平」定义在群体基准率不同时无法同时满足(如校准与群体间平衡),是讨论算法歧视时常引用的不可能性结果。

奠基之作难度 3 / 5约 60 分钟经得起时间
读之前最好知道
1概率分类与风险评分
2校准(calibration)
3假阳性率与假阴性率
4基准率(base rate)与条件概率

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
推荐bbig3831 · 2017-04

“I'd recommend the articles for a better explanation.”

推荐阅读这篇论文,它说明各种公平性定义无法同时成立
Courts are using risk-assessment software to sentence criminals原评论 ↗
推荐bobcostas55 · 2018-06

“I recommend _The impossibility of “fairness”: a generalized impossibility result for decisions_[0] and _Inherent Trade-Offs in the Fair Determination of Risk Scores_[1]”

把它列为公平性与风险评分取舍的推荐读物
AI at Google: our principles原评论 ↗
引用yummyfajitas · 2016-11

“The conclusion there is simply a theorem: it's mathematically impossible to be both well calibrated”

引用其定理:良好校准与种族平衡不可能同时成立
Facebook is harming our democracy原评论 ↗
引用aub3bhat · 2016-11

“It turns out there is a beautiful mathematical result that shows that the different definitions of "fairness" are inherently incompatible with each other and cannot be satisfied at the same time.”

拿它说明不同的"公平"定义在数学上互不相容、无法同时满足
A machine learning approach to identifying thought markers of suicidal subjects原评论 ↗
引用yummyfajitas · 2016-12

“there are impossibility theorems that say you can't meet everyone's definition of "nondiscriminatory" except in a few trivial cases.”

引用不可能性定理:除少数平凡情形外,无法满足所有人对"无歧视"的定义
Unexpected sources of bias in artificial intelligence原评论 ↗
引用sanxiyn · 2018-01

“We actually have a mathematical proof (which is quite simple) why this is impossible.”

指出已有简单的数学证明说明这种公平不可能实现
Algorithmic decision making and the cost of fairness (2017)原评论 ↗
看全部 8 条 →
引用lalaland1125 · 2018-09

“Of course, defining "racism" itself gets extremely tricky because many definitions of racism are mutually contradictory.”

拿它佐证算法里对"种族主义"的各种定义彼此矛盾
Making a racist AI without really trying (2017)原评论 ↗
引用Legend2440 · 2024-06

“that cannot possibly be fulfilled at the same time if there is any difference between the groups.”

拿它说明只要群体间有差异,这些公平标准就不可能同时满足
Mozilla.ai did what? When silliness goes dangerous原评论 ↗
被提交为 story · 1 次