Project Grant 2610282

The University of Pennsylvania received a $200,000 Project Grant from the National Science Foundation's Division of Mathematical Sciences (Mathematical and Physical Sciences program, CFDA 47.049) awarded July 1, 2026, for completion by June 30, 2029. The project develops distribution-free statistical inference methods to assess the reliability and trustworthiness of artificial intelligence (AI) systems when their predictions are used to make consequential decisions in adaptive pipelines. The research addresses critical gaps in uncertainty quantification for AI applications by drawing on conformal prediction, selective inference, multiple testing, and permutation methods to provide finite-sample, model-agnostic validity guarantees that remain robust when predictions are selected adaptively or applied iteratively.

The deliverables include set-level predictive inference methods incorporating false discovery rate control, family-wise error rate control, and global null testing capabilities designed for use across multiple unlabeled instances. Beyond core research activities, the project will generate publicly available software tools, statistical benchmarks, and educational materials to support reproducible research and safer AI deployment. The work emphasizes broader impacts through graduate and undergraduate student training in trustworthy AI and responsible data science, with intentional efforts to broaden participation in these fields. Applications of the statistical quality-control layer are expected to improve reproducibility and efficiency in biomedical science, drug discovery, and other data-intensive fields where experimental costs are high and decision errors carry significant consequences.