Scientific Machine Learning and Artificial Intelligence: Uncertainty Quantification

archived

In support of the Executive Order on Maintaining American Leadership in Artificial Intelligence, the DOE Artificial Intelligence (AI) Program and DOE SC program in Advanced Scientific Computing Research (ASCR) hereby announce their interest in the co-design of learning systems and AI environments that significantly advance the field of AI for public benefit within DOE’s Congressionally-authorized mission-space. The principal focus of this FOA is on Uncertainty Quantification (UQ) for AI validation and prediction. Foundational research is needed for strengthening the mathematical and statistical basis of validating machine learning and AI predictions from data generated by the Office of Science’s user facilities and scientific simulations. A critical open question for scientific machine learning (SciML) is: How do we make reliable predictions and uncertainty estimates from machine learning and AI models? Predictions can be greatly improved by including input uncertainties and insights from model discrepancies. Research advances will be needed in methods that incorporate mathematical, statistical, scientific, and engineering principles for uncertainty estimates in extrapolative predictions. Furthermore, extensive literature in statistics can be leveraged for improving the model validation process. Advances in UQ will greatly enhance the mathematical and scientific computing foundations for accelerated research insights from SciML and AI.

High