2025

Halonen, Vilho; Pölönen, Ilkka; Wolfmayr, Monika

We create and test a neural network which quantifies uncertainty errors in the Poisson equation generated by an uncertain source term. The neural networks performance is compared to a Monte Carlo approximation and analytically derived bounds. We find that with a suitable dataset the neural network can learn this task well enough to be considered as an alternative for Monte Carlo and analytical methods.