Project Grant 2212263

This $600,000 project grant from the National Science Foundation's Computer and Information Science and Engineering program will support research at Stanford University toward developing a mathematical foundation for deep reinforcement learning. Over four years, the grant will fund three research thrusts investigating the types of guarantees achievable by reinforcement learning policies under different problem structures and increasing neural network complexity. The researchers will also design computationally efficient algorithms by leveraging tools from optimization and control theories. As part of the statutory broader impacts review criteria, the grant incorporates education and outreach plans including mentoring graduate and undergraduate students, developing new courses and materials, and organizing workshops. This award is aligned with the NSF program's goals of supporting investigator-initiated research and advancing the development of cyberinfrastructure to accelerate discovery and innovation in computing and information sciences.