Constrained Extreme Gradient Boosting for Adapting Reduced-Order Models

Researchers propose Constrained Extreme Gradient Boosting (cXGBoost), an ensemble learning framework that predicts Proper Orthogonal Decomposition bases as functions of system parameters by mapping Grassmann manifold subspace representations to Euclidean space for gradient boosting regression, enabling adaptive reduced-order models that maintain accuracy across parameter variations in expensive simulations like CFD and finite element analysis.