This is the authors' abstract. We don't add key points for this paper.
View Video Presentation: https://doi.org/10.2514/6.2023-0902.vid Emulator embedded neural networks are a type of physics informed neural network which can leverage multi-fidelity data sources for efficient design exploration of aerospace engineering systems. Multiple realizations of the neural network models can be trained with different random initializations. The ensemble of model realizations is used to assess epistemic modeling uncertainty, leading to goal-oriented adaptive learning for more useful predictions of system response. While emulator embedded neural networks are scalable to high dimensional problems, they can require significant training costs, especially for ensembles of models, depending on the network structure and size of the data set. This can pose a computational challenge when conducting adaptive learning and design exploration of an aerospace system, especially if parallel training of multiple models is not implemented. In this work, a new type of emulator embedded neural network is developed using the rapid neural network paradigm. This involves initializing a neural network with randomly chosen weights and biases. Only the last layer's weights and biases are adjusted, using linear regression techniques such as Ridge regression or Lasso. This skips the lengthy gradient descent training process entirely, resulting in near-instantaneous creation of the emulator embedded neural network model. The proposed method is demonstrated on multiple analytical examples, as well as a preliminary aerospace design study of a generic hypersonic vehicle.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Beachy et al. (2023) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: