orthogonal discrepancy kernel
Orthogonal Discrepancy Kernels for Learning with Partial Physics
Manna, Swapnil, Rogers, Timothy J., Bull, Lawrence
We introduce a semi-parametric framework for nonlinear system identification, which decouples discrepancy functions from physics-based components. Orthogonal Gaussian process regression balances sparse parameter selection (the white box) with discrepancy learning (the black box) to produce interpretable models from incomplete physics.