1D Visualization of a Successive Surrogate Modeling Optimization Method скачать в хорошем качестве

1D Visualization of a Successive Surrogate Modeling Optimization Method

Surrogate modeling (or meta-modeling) is a strategy often used in engineering design optimization when objective or constraint functions require evaluation of a computationally-expensive simulation. Successive (or adaptive) surrogate modeling methods gradually increase surrogate model accuracy in areas of strategic importance to help improve the chances of finding an accurate global optimum, while reducing the number of expensive simulations required. This video demonstrates an adaptive surrogate modeling approach applied to a multi-modal function of a single variable to support effective visualization of the method. The surrogate model is based on a neural network, and model parameters (# of neurons, domain size and location) adapt during the process based on heuristic rules.