A new arXiv preprint announces work on structure-aware spatial kernels for Gaussian process distributed port-Hamiltonian systems (GP-dPHS). These systems combine data-driven modeling with the preservation of physical structure, making them useful for distributed dynamical systems.
The abstract notes that standard stationary kernels can be inefficient for such settings. The paper appears to address this limitation by introducing spatial kernels that are aware of the system's structure, though the specific inefficiencies are not detailed in the available abstract text.
Because only the announcement is available, the full scope of the proposed method remains unclear. The contribution, as stated, is a step toward more efficient kernel design within the GP-dPHS framework.