Probabilistic neural networks for fluid flow surrogate modeling and data recovery

Artificial neural networks (ANNs) have recently been applied to several fluid dynamics applications. However, there are a very limited number of studies that assess the fidelity of ANN deployments, a function of algorithm choice and training data quality, by quantifying uncertainties in predictions. This diminishes their utility for practical modeling requirements. In an effort to address this, a probabilistic NN that provides confidence intervals for its predictions in a computationally effective manner is used. This approach is demonstrated in surrogate modeling and flow reconstruction tasks with promising results.