Project Grant 2339395

This National Science Foundation (NSF) Computer and Information Science and Engineering (CISE) Program Project Grant (CFDA 47.070) of $151,593 awarded to Cornell University on March 1, 2024 will support the development of new reinforcement learning (RL) algorithms that can learn efficiently and reliably from limited training data. The key products of this 5-year project will be RL algorithms that can be safely deployed in real-world applications like autonomous driving and generative AI where data is expensive to collect and reliability is critical. The research aims to advance RL techniques by developing risk-averse RL algorithms, leveraging problem-specific structures for improved sample efficiency, and creating new RL algorithms that can incorporate rich human feedback beyond scalar rewards. The project will focus on deploying the developed RL technologies to database query optimization and optimizing generative models like large language models and diffusion models.