Orthogonal Discrepancy Kernels for Learning with Partial Physics
arXiv:2606.211993.1
Predicted impact top 89% in ML · last 90 daysOriginality Incremental advance
AI Analysis
It addresses the problem of learning accurate models when physics knowledge is incomplete, benefiting system identification practitioners.
The paper introduces a semi-parametric framework for nonlinear system identification that decouples discrepancy functions from physics-based components, using orthogonal Gaussian process regression to produce interpretable models from incomplete physics.
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.