Multi-dimensional Domain Generalization with Low-Rank Structures

Explore a novel approach to domain generalization in linear regression models through this 58-minute conference talk. Delve into the challenges of making statistical inferences about underrepresented populations, particularly in health-related studies. Learn how organizing model parameters for subpopulations into a tensor and studying structured tensor completion can achieve robust domain generalization. Discover how this innovative method leverages group label structures to produce more reliable and interpretable results for subpopulations with limited or no available data. Gain insights into the rigorous theoretical guarantees and minimax optimality of the proposed approach, addressing the limitations of conventional statistical and machine learning methods that assume identical distribution between test and training data.

Sai Li: Multi-dimensional domain generalization with low-rank structures #ICBS2024