$PINN - a Domain Decomposition Method for Bayesian Physics-Informed Neural Networks
(2026) AI&PDE: ICLR 2026 Workshop on AI and Partial Differential Equations — Location: Rio de Janeiro, Brazil (23.April.2026)
Files
ICLR_AI_PDE_PINN_a_Domain_Decomposition_method.pdf
Open Access - Adobe PDF
- 1.41 MB
Details
- Authors
- Abstract
- Physics-Informed Neural Networks (PINNs) are a novel computational approach for solving partial differential equations (PDEs) with noisy and sparse initial and boundary data. However, efficient quantification of epistemic and aleatoric uncertainties in big multi-scale problems remains challenging. We propose $PINN, a novel method of computing global uncertainty in PDEs using a Bayesian framework , by combining local Bayesian PINNs (BPINNs) with domain decomposition. The solution continuity across subdomains is obtained by imposing flux continuity across the interface of neighboring subdomains. Although we have adopted conservative PINNs (cPINNs), the method can be seamlessly extended to other domain decomposition techniques. The results show that the proposed method recovers the global uncertainty by computing the local uncertainty exactly more efficiently, as the uncertainty in each subdomain can be computed concurrently. The robustness of $PINN is verified by adding uncorrelated random noise to the training data up to 15% and testing for different domain sizes.
- Affiliations
Citations
Vicens Figueres, J., Vanderhaeghen, J., Bragone, F., Morozovska, K., & Shukla, K. (2026). $PINN - a Domain Decomposition Method for Bayesian Physics-Informed Neural Networks. AI&PDE: ICLR 2026 Workshop on AI and Partial Differential Equations, Rio de Janeiro, Brazil. https://doi.org/10.2139/ssrn.5290827