Project Grant 2212418

This National Science Foundation project grant of $600,000 supports research establishing conditional transport as a new statistical distance measure between probability distributions to address limitations of existing methods. Funded under the Computer and Information Science and Engineering program, the award to the University of Texas at Austin from October 2022 to September 2026 will develop a new distribution-based machine learning framework and efficient approximation algorithms. Key products include advancing distribution-based machine learning and artificial intelligence fundamental research through theoretical understanding of conditional transport. The framework is intended to provide desired model representation power and learning stability. Maximum likelihood estimation, Bayesian inference, and entropy regularized optimal transport will be revisited based on conditional transport. New machine learning models and inference algorithms for deep generative modeling, contrastive representation learning, and deep reinforcement learning are expected to advance the state of the art. Inverse materials design and multi-omics data analysis applications will demonstrate advantages of conditional transport-based methods for critical decision making requiring reliable uncertainty quantification.