ATD: Privacy-preserving Federated Neural Operators for Human Mobility Prediction on Cross-Domain Infrastructures

Lehigh University, Bethlehem PA

Investigators

Abstract

This project aims to revolutionize the understanding and prediction of human mobility by leveraging the vast data generated through advanced urban infrastructures and commercial systems. Understanding real-time population density and how human mobility is affected by crises can significantly improve early interventions and response strategies. By analyzing large-scale human mobility data from sources like smartphones, payment systems, cellular towers, and transportation networks, this project seeks to overcome the limitations of existing research, which often relies on biased, incomplete, or noisy data from single domains. This project aligns with NSF’s mission to promote the progress of science by providing mathematical and statistical algorithms for the analysis of large spatiotemporal datasets with applications to quantitative models of human dynamics. Additionally, this work will include educational and outreach activities, promoting diversity and inclusion within the scientific community and beyond. The project will address two fundamental challenges: spatiotemporal data heterogeneity across various domains and privacy concerns during cross-domain collaboration. To achieve this, the investigators will design domain-invariant spatiotemporal modeling techniques for human mobility prediction, develop new frameworks for collaborative learning across various domains without compromising data privacy, and create algorithms for threat detection, targeting, and mobility prediction during crises. This research will utilize real-world data from two cities, incorporating 13 types of mobility data. The intellectual merit of this project includes the development of a spatiotemporal nonlocal neural operator model, a unified framework for federated unsupervised graph learning, and domain-aware anomaly detection models for threat scenarios. Beyond human mobility prediction, the technological advancements will contribute to a wide range of spatiotemporal modeling applications. The methods and tools developed will be made available as open-source software, and the investigators will conduct various dissemination activities to share the findings with both academic and broader communities. The educational framework integrates outreach activities to ensure that diverse groups benefit from the research. 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.