Hello,

I'd like to share an issue about the coefficients data of EBF-NN based surrogate model.

First, an initial surrogate model (based on EBF-NN) is performed with sampling points, and the coefficients data are exported after completing the initialization.

Then, the coefficients data of EBF-NN is used to recreate a surrogate model, the error analysis results (using the same separate data file) are different to the initial model. So I guess there may be a mistake on the coefficients data.

If an optimization component is used in consociation with the surrogate model (ebf-nn based), the predicted points are calculated with using these coefficients data, thus the optimization results will be false.

Isight ​​​​​​​