SingleTaskMultiFidelityGP is a BoTorch Gaussian-process class for learning one objective from observations paired with fidelity values. Its multifidelity kernel transfers information between fidelity levels according to their correlation, reducing uncertainty even when observations come from a cheaper, lower-fidelity process. section 9.2.2
Predict at high fidelity
Evaluate the trained model on a grid paired with ones as the high-fidelity value, then read its mean and confidence region. Low-fidelity observations can still narrow these predictions when the kernel correlation is substantial.
