Federal Grant Award Summary

The Johns Hopkins University received a $188,168 Project Grant from the National Science Foundation's Division of Computing and Communication Foundations under the Computer and Information Science and Engineering (CISE) program (CFDA 47.070), effective July 1, 2026 through July 31, 2027. This collaborative research award supports the development of comprehensive theory, algorithms, and applications for large-scale bilevel optimization—a nested optimization structure increasingly prevalent in emerging machine learning fields including meta-learning, hyperparameter selection, continual learning, and adversarial learning.

The project delivers research outputs across three integrated thrusts: (1) development of fast, scalable Hessian-free bilevel algorithms with convergence rate guarantees using momentum-based methods and finite-difference matrix-vector estimation; (2) advancement of primal-dual, primal, and pessimistic bilevel methods with convergence analysis for cases lacking unique lower-level solutions; and (3) design and analysis of algorithms for solving bilevel problems on non-linear manifolds. The outcomes will provide theoretical foundations and computational methods applicable to large-scale nested optimization problems across information science, signal processing, communications, statistics, and machine learning disciplines, benefiting researchers in academia, government laboratories, and industry.