CAREER: New Frameworks for Ethical Statistical Learning: Algorithmic Fairness and Privacy

随着数据科学和机器学习在我们日常生活的许多方面（如医疗保健、金融、教育和法律）产生前所未有的影响，迫切需要设计符合道德的统计学习算法，以兼顾公平和隐私。该项目应对将道德原则纳入统计学习结构的挑战。该方法通过增强统计算法以公平地执行来优先考虑公平性，特别是在样本量有限以及敏感属性受到法律的或社会规范限制的情况下。与此同时，该项目通过开发一个通用框架来解决隐私问题，该框架用于研究随着生成人工智能的进步而出现的新隐私约束下的隐私-准确性权衡。这项工作的实际结果是将这些方法应用于生物医学领域，同时发布开放源码软件，扩大影响，鼓励在各个领域的统计学习中采取道德做法。该项目旨在促进公平和隐私的数据处理，并为学生提供研究培训机会。该项目的研究目标是为道德机器学习制定严格的统计框架，重点关注算法公平性和数据隐私。更具体地说，该项目将：（1）开发创新的统计方法，以有限样本和无分布的方式确保公平性;（2）设计算法，确保公平性，同时遵守对敏感数据的社会和法律的限制;（3）建立新的框架，以阐明生成人工智能中统计准确性和新隐私概念之间的权衡，包括机器学习和版权保护。总的来说，这项研究的成果将为伦理统计学习奠定坚实的基础，并为新的理论理解和实践方法的发展提供启发，同时提供算法公平性和隐私保障。该奖项反映了NSF的法定使命，并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，被认为值得支持。

With the unprecedented impact of data science and machine learning in many aspects of our daily lives, such as healthcare, finance, education, and law, there is an urgent need to design ethical statistical learning algorithms that account for fairness and privacy. This project tackles the challenge of integrating ethical principles into the fabric of statistical learning. The approach prioritizes fairness by enhancing statistical algorithms to perform equitably, particularly in scenarios with limited sample sizes and where sensitive attributes are restricted by legal or societal norms. In parallel, this project addresses privacy by developing a general framework for studying the privacy-accuracy trade-off under new privacy constraints emerging with the advances in generative AI. The practical upshot of this work is the application of these methods to biomedical fields, accompanied by the release of open-source software, broadening the impact and encouraging ethical practices in statistical learning across various domains. This project promotes equitable and private data handling and provides research training opportunities to students.The research objective of this project is to develop rigorous statistical frameworks for ethical machine learning, with a focus on algorithmic fairness and data privacy. More specifically, the project will: (1) develop innovative statistical methods that ensure fairness in a finite-sample and distribution-free manner; (2) design algorithms that ensure fairness while complying with societal and legal constraints on sensitive data; (3) establish new frameworks to elucidate the trade-off between statistical accuracy and new privacy concepts in generative AI, including machine unlearning and copyright protection. Taken together, the outcome of this research will build a firm foundation of ethical statistical learning and shed light on the development of new theoretical understanding and practical methodology with algorithmic fairness and privacy guarantees.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.