EAGER:AI-DCL:Capture, Explain and Negotiate the Inherent Trade-offs in Machine Learning Algorithms

EAGER：AI-DCL：捕获、解释和协商机器学习算法中固有的权衡

基本信息

批准号：
1927166
负责人：
Haiyi Zhu
金额：
$ 29.57万
依托单位：
University of Minnesota-Twin Cities
依托单位国家：
美国
项目类别：
Standard Grant
财政年份：
2019
资助国家：
美国
起止时间：
2019-10-01 至 2019-12-31
项目状态：
已结题

来源：
https://www.nsf.gov/awardsearch/showAward?AWD_ID=1927166&HistoricalAwards=false
关键词：
EAGER AI DCL Capture Explain

项目摘要

An outstanding issue with machine learning based decision-making algorithms is the inherent trade-offs between different system criteria. There is an emerging body of literature demonstrating trade-offs between fairness and accuracy, and between different fairness notions. By improving fairness, overall accuracy might decrease. Furthermore, different fairness notions are not compatible with each other: well-established results show that common statistical fairness notions are often mutually exclusive. Accurate understanding of such trade-offs is critical for stakeholders and practitioners to appropriately use these machine learning methods. The focus of this project is to take an interdisciplinary approach to study, explain, and address the inherent trade-offs between different system criteria in machine learning-based decision-making.The researchers will develop methods to capture trade-offs between different system criteria in machine learning algorithms. They will develop visualizations and interactive interfaces to explain the trade-offs between the models to the stakeholders. Finally, the project team will explore social and technical innovations that let stakeholders navigate and negotiate the fundamental trade-offs between different system criteria. The project will disseminate the new knowledge in part through a MOOC offering.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

基于机器学习的决策算法的一个杰出问题是不同系统标准之间的固有权衡。有一大批文学体系，表明公平与准确性之间以及不同的公平概念之间的权衡。通过提高公平性，总体准确性可能会降低。此外，不同的公平概念彼此不兼容：成熟的结果表明，常见的统计公平概念通常是相互排斥的。对这种权衡的准确理解对于利益相关者和从业者适当使用这些机器学习方法至关重要。该项目的重点是采用跨学科的方法来研究，解释和解决基于机器学习的决策中不同系统标准之间的固有权衡。研究人员将开发方法来捕获机器学习算法中不同系统标准之间的权衡。他们将开发可视化和交互式接口，以向利益相关者解释模型之间的权衡。最后，项目团队将探索社会和技术创新，使利益相关者在不同系统标准之间进行基本权衡。该项目将部分通过MOOC发行来传播新知识。该奖项反映了NSF的法定任务，并被认为是值得通过基金会的知识分子优点和更广泛的影响评估标准来评估值得支持的。