Training and trust

How a surrogate is trained, how to validate it honestly on unseen designs, and where it should not be trusted.

• Training: Fit the weights once on the training simulations and watch the validation curve.
• Validation: Compare surrogate and solver on designs the network never saw, inside and outside the training range.
• Limits and trust: A surrogate is only as good as its data: know the domain, quantify error, verify the winner.