Neural network approximator with novel learning scheme for design optimization with variable complexity data

A novel learning scheme for training neural networks is proposed. The trained network is then used for function approximation during the numerical optimization process. The learning scheme trains the network with data of varying complexity, including data that have only zeroth-order information, and when the data include first-order information (gradients of responses with respect to the design parameters) it uses the combined information (response values and its gradients) for a better approximation. The learning scheme and its function approximation capability for design optimization are demonstrated on two realistic examples.