Project Grant 2610298

The National Science Foundation Division of Mathematical Sciences awarded the University of South Carolina $159,813 on September 1, 2026, under the Mathematical and Physical Sciences program (CFDA 47.049) to develop generative bootstrapping methods for uncertainty quantification in massive dependent data.

The project combines approximate inference for dependent data with generative artificial intelligence tools to enable faster, more reliable, and scalable uncertainty assessment for applications including transportation systems, healthcare, manufacturing, remote sensing, and digital twins. Standard approaches for constructing large-sample confidence intervals and prediction intervals are computationally expensive and difficult to scale when applied to large dependent data. The research will introduce a computationally efficient generative bootstrapping framework that replaces traditional repeated resampling and re-estimation procedures with surrogate likelihood methods such as the Vecchia approximation and composite likelihood, reducing computational burden particularly on resource-constrained hardware.

The award supports development of open-access educational materials, public software, and training for undergraduate and graduate students in modern statistics, data science, and AI. Performance occurs in Columbia, South Carolina, with a period of performance from September 1, 2026, through August 31, 2029. This is a collaborative research project grant.