Fig. 3: Comparison of physics-informed and black-box modeling of Eq. (3). | npj Computational Materials

Fig. 3: Comparison of physics-informed and black-box modeling of Eq. (3).

From: A physics informed bayesian optimization approach for material design: application to NiTi shape memory alloys

Fig. 3

All Gaussian processes are initialized with identical training data at different length scales to examine the impact of smoothness as well as utilization of physics-infused kernel. As shown, using a physics-infused kernel helps to catch the correct form of function variability and extremum approximations. When the function displays irregular variability between observations, black-box modeling is insufficient to accurately represent the function’s true form.

Back to article page