Practice / Training and validation

Training

Fit the weights once on the training simulations and watch the validation curve.

Training draws minibatches of points from training designs, compares predicted and simulated fields and updates the weights; a held-out validation set measures generalisation.

Related

• Sampling the design space: Spread a limited simulation budget evenly over the designs you care about.
• Validation: Compare surrogate and solver on designs the network never saw, inside and outside the training range.
• Technique library: Thirty techniques that make surrogates work in practice, from basics to the frontier.