Federal Project Grant Award Summary

Arizona State University's Office of Research and Sponsored Projects Administration received a $300,000 Project Grant awarded September 1, 2026, through the National Science Foundation's Division of Computing and Communication Foundations under the Computer and Information Science and Engineering program (CFDA 47.070). The three-year collaborative research project, scheduled for completion by August 31, 2029, focuses on developing uncertainty quantification methods for large language models (LLMs) to improve their reliability in high-stakes applications such as transportation planning and healthcare. The research addresses a critical gap: while LLMs are increasingly deployed to support multi-step reasoning and decision-making, they frequently produce erroneous conclusions while expressing them with high confidence, creating substantial risks in consequential domains.

The project delivers three primary research thrusts. First, the team will develop accurate and efficient methods to quantify uncertainty in multi-step reasoning by representing model reasoning as graphs and applying graph augmentation techniques to capture structural inconsistencies and intermediate reasoning ambiguities. Second, the project will create a unified framework that integrates distinct uncertainty forms—including semantic ambiguity, structural reasoning errors, and inconsistent world knowledge—into a comprehensive characterization of model behavior. Third, the team will introduce an uncertainty-aware adaptation method that attributes uncertainty to fine-grained reasoning steps and incorporates these signals into training procedures including reinforcement learning from human feedback, direct preference optimization, and group relative policy optimization. These deliverables aim to accelerate the safe adoption of artificial intelligence technologies in high-risk domains requiring stable and reliable performance.