CAREER: Foundations of Collaborative Machine Learning

机器学习的最新进展依赖于收集大量数据和在集中式云中学习大量模型。然而，集中化方法的过度存储和计算需求，以及共享私人数据方面的监管挑战，使这一范例的实用性受到质疑。协作机器学习是最近的一种解决这些问题的替代范式，它通过协作开发算法来解决这些问题，而不需要交换或集中数据。例如，不同地理位置分布的医院，每一家都拥有有限的患者数据，可以合作开发预测算法，以改进诊断和治疗，而不是单独完成的任务。能否充分发挥协作学习的潜力，很大程度上取决于能否鼓励大量个人或公司共享他们的私人数据和资源。为了实现这一目标，该职业奖提供了一种交叉方法来开发基于理论的协作算法，通过联合解决各种计算、统计、系统和博弈论挑战，促进在资源限制下从碎片化、异质的私有数据中进行最佳学习。通过促进稳定和公平的生态系统，使所有参与者受益并留住所有参与者，而不施加严格的数据和资源限制，该项目的成果承诺使数据驱动的智能系统在各种应用领域更加有效、个性化和健壮，如个性化医疗保健、精准农业和教育。为了促进健康的数据和计算生态系统，并能够在资源约束下优化使用分布式异质数据，该项目提供了一种交叉方法，以严格解决计算、统计和博弈论方面的挑战。在实践方面，它引入了多元化的学习范式，并开发了分布式算法，这些算法认识到统计异质性，并限于满足可用资源的学习模型。在统计方面，该项目侧重于建立泛化保障和理解信息论的权衡，为协同进步和洞察提供机会。该项目在协作学习和聚合游戏之间建立了密切的联系，并利用发展的理论和算法调查来回答与平衡、激励、公平和稳定有关的问题，以促进健康的生态系统。该项目提出的交叉和统一的研究，创造了必要的联系，并促进了新的变革性方法，而不是通过个别学科的努力开发出来的。这项研究将通过举办研讨会、指导学生和开发课程与教育相结合。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Recent advances in machine learning rely on collecting an enormous amount of data and learning immense models in a centralized cloud. However, the excessive storage and computational needs of centralized approaches, alongside regulatory challenges in sharing private data, put the utility of this paradigm in doubt. Collaborative machine learning is a recent alternative paradigm to tackle these issues by developing algorithms collaboratively without exchanging or centralizing the data. For example, different geographically distributed hospitals, each being in possession of limited patients’ data, may collaboratively develop predictive algorithms to improve diagnostics and treatment beyond what could be accomplished alone. Unlocking the full potential of collaborative learning strongly depends on the ability to encourage a large pool of individuals or corporations to share their private data and resources. Towards this aim, this CAREER award offers an intersectional approach to develop theoretically-grounded collaborative algorithms to facilitate learning optimally from fragmented, heterogeneous private data under resource constraints by jointly addressing various computational, statistical, systems, and game-theoretic challenges. By promoting a stable and fair ecosystem to benefit and retain all participants, without imposing stringent data and resource constraints, this project’s outcomes promise to make data-driven intelligent systems more effective, personalized, and robust in a myriad of application domains, such as personalized healthcare, precision agriculture, and education.To promote a healthy data and compute ecosystem and enable optimal use of distributed heterogeneous data under resource constraints, this project offers an intersectional approach to rigorously address computational, statistical, and game-theoretic challenges. On the practical side, it introduces a pluralistic learning paradigm and develops distributed algorithms that are cognizant of statistical heterogeneity and confined to learning models that meet available resources. On the statistical side, the project focuses on establishing generalization guarantees and understanding information-theoretic tradeoffs, providing an opportunity for synergistic advancements and insights. The project makes an intimate connection between collaborative learning and aggregative games and leverages developed theoretical and algorithmic investigations to answer questions related to equilibrium, incentivization, fairness, and stability to promote a healthy ecosystem. The intersectional and unified study the project proposes, creates essential connections, and fosters new transformative methods not developed by efforts within the individual disciplines. The research will be integrated with education through hosting workshops, mentoring students, and developing courses.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.