CAREER: Advancing Trustworthy Machine Learning for Distributed Scientific Data Analytics

University Of Delaware, Newark DE

Investigators

Abstract

This NSF CAREER project presents a comprehensive research and education strategy aimed at creating a trustworthy optimization toolbox for geo-distributed scientific data analytics. The toolbox is designed to address a critical gap in current artificial intelligence (AI)/machine learning (ML) practices, where predictive models are typically trained on historical and/or regional data. This approach is inadequate for capturing the full spectrum of complex and evolving phenomena, such as extreme weather events and climate change. The project focuses on innovating optimization methods that improve the robustness of predictions, the reliability of explanations, and the scalability of privacy protections. These innovations are essential for increasing the trustworthiness of AI/ML systems when dealing with rare or previously unseen scenarios, thereby enabling decisions that can be trusted by domain experts. The success of this project is expected to establish a solid foundation for recognizing AI/ML as a legitimate scientific methodology for high-stake, safety-critical applications. The project holds substantial merit by pursuing three research aims. Aim-1 bridges the long-standing gap between data topology and robust optimization, resulting in a new topological robust optimization framework for out-of-distribution generalization. Aim-2 revolutionizes the use of explainable machine learning by interpreting model predictions at data, knowledge, and concept levels, facilitating interactive scientific knowledge discovery. Aim-3 addresses the pressing need for trustworthy collaborative learning in out-of-federation scenarios, ensuring scalable and explainable data protection. To validate these advancements, the project uses standard benchmarks alongside newly curated AI-ready datasets in three applications: flood water mapping, seafloor characterization, and global urbanization forecasting. To maximize societal impact, this project integrates research findings at all educational levels through interdisciplinary collaborations. Initiatives include promoting undergraduate AI4Sciences research, developing interdisciplinary curricula, supporting underrepresented groups in computer science, and enhancing outreach efforts at K-12 and community levels. This goal is to cultivate a diverse and inclusive STEM workforce, ensuring the benefits of this research reach broad society. This project is jointly funded by Information Integration and Informatics and the Established Program to Stimulate Competitive Research (EPSCoR). 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.