CAREER: Information, Algorithms, and Learning

该奖项资助经济理论和计算机科学交叉领域的研究。该项目的动机是越来越多地使用机器学习算法来指导决策，例如谁应该获得保释或获得贷款。该项目将有助于更好地了解此类算法方法的机会和危险。主要研究者（PI）计划分两步工作。首先，PI将从理论上对算法预测的福利影响进行建模，重点关注算法的“公平性”（即，算法的错误是否由一个社会群体相对于另一个社会群体不成比例地承担）以及算法的准确性。其次，PI将提出新的方法来理解黑箱方法的预测与经济模型的预测有何不同，特别关注这些预测在新的未知领域的推广情况。这项研究将在经济理论和经济数据解释的新方法上取得新的成果。该奖项还资助了一所大学经济学和计算机科学相结合的新课程开发。第一个项目提供了一个概念框架的算法决策和“公平性准确性”帕累托边界如何依赖于算法的输入特性。第二个项目提出了评估经济模式的新措施，包括衡量经济模式的限制性和衡量经济模式的可移植性。最后一个项目提出了一个信息流在社交媒体平台上的模型，当社会规范，管理可接受的表达是在流动。该项目的重点是表征表达的观点何时以及如何系统地不同于意见的真实分布。该奖项反映了NSF的法定使命，并被认为值得通过使用基金会的知识价值和更广泛的影响审查标准进行评估来支持。

This award funds research on topics at the intersection of economic theory and computer science. This project is motivated by the growing use of machine learning algorithms to guide decisions such as who should receive bail or receive a loan. This project will contribute towards a better understanding of the opportunities and the dangers of such algorithmic approaches. The principal investigator (PI) plans to work in two steps. First, the PI will theoretically model the welfare implications of algorithmic predictions, focusing on tradeoffs between goals such as the “fairness” of the algorithm (i.e., whether the algorithm’s errors are disproportionately borne by one social group over another) and the accuracy of the algorithm. Second, the PI will propose new methods for understanding how predictions made by black box methods differ from those made by economic models, focusing in particular on how well these predictions generalize to new unseen domains. This research will develop new results in economic theory and new methodologies for the interpretation of economic data. The award also funds new curriculum development for a university major combining Economics and Computer Science.The overall project has three components. The first project provides a conceptual framework for algorithmic decision-making and a characterization of how the “fairness-accuracy” Pareto frontier depends on the inputs to the algorithm. The second project proposes new measures for evaluating economic models, including a measure for the restrictiveness of the economic model and a measure for the portability of the economic model. The final project proposes a model of information flow on social media platforms when the social norms that govern acceptable expression are in flux. This project focuses on characterizing when and how expressed views systematically differ from the true distribution of opinions.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.