Federal Grant Award Summary

The National Science Foundation's Division of Mathematical Sciences awarded $133,285.00 to the University of California, Berkeley under the Mathematical and Physical Sciences (CFDA 47.049) program for a collaborative research project on calibrated hypothesis testing, effective June 1, 2026 through May 31, 2029. The project develops statistical methods and theory to improve the reliability of large-scale scientific inference by providing individual research findings with interpretable, rigorous error probability guarantees. Rather than controlling only average error rates across entire lists of discoveries, the research will create methodologies that distinguish strong evidence from borderline evidence through calibrated assessment, addressing critical gaps in current statistical tools used across metascience and artificial intelligence (AI) safety applications.

Deliverables from this award include the development of theoretical and methodological frameworks combining empirical Bayes estimation with frequentist finite-sample guarantees, extension of local and boundary false discovery rate methods beyond independent p-value settings, and variable selection applications using knockoff and sign-symmetric statistics. Additional outputs include training of graduate researchers, creation of instructional materials, and release of open-source software to support more reproducible science and safer data-driven AI/machine learning (ML) systems. This work directly advances the National Science Foundation's objectives to strengthen the nation's scientific enterprise while enhancing understanding of fundamental methodological challenges in modern large-scale data analysis.