Project Grant 2323112

This Project Grant award from the National Science Foundation (NSF) Office of Advanced Cyberinfrastructure under the CFDA program "Computer and Information Science and Engineering" provides $150,000 to Duke University to conduct collaborative research on developing efficient and provable algorithms for approximate sampling-based exploration strategies in sequential decision-making problems. The key objectives are to unify exploration strategies across different applications like recommendation systems, robotics, games, and personalized healthcare, and to provide research training for graduate students. The award will fund three main tasks: (1) developing fast approximate sampling-based exploration strategies for contextual bandit problems; (2) implementing and generalizing these exploration algorithms to more complex sequential decision-making applications using deep neural networks; and (3) establishing efficient and provably effective exploration strategies for reinforcement learning problems. The research outcomes will be translated into accessible open-source software and course materials to benefit research, education, and society. No sub-awards are planned under this grant.