Figure 1.

Two alternative GP regressions for the high-fidelity model f_h(t) = sin²(8πt) (the highly oscillatory green curve in (a)). The blue curve visually suggests a smooth two-dimensional manifold g(t, f_l). If, on the other hand, the high-fidelity function is projected onto f_l(t), we can see it is a simple quadratic. (b,c) The prediction results with 2 s.d. (dashed line) for the high-fidelity function by GP regression with only t and with only f_l(t), respectively. We employ 15 high-fidelity data points and 100 uniformly distributed low-fidelity data points for the regression. (a) A dependency between f_l(t) and f_h(t) with t, (b) GP with {t}, (c) GP with {f_l(t)}. (Online version in colour.)