Our model can also serve as a plug-and-play replacement for shallow manifold GPs in geometry-aware Bayesian optimisation. This can be especially useful for complex target functions, as we demonstrate experimentally. 6/n

Comments