We are pleased about our contribution to ICINCO 2024 in Porto. Congratulations to the authors!
Domain-decoupled Physics-informed Neural Networks with Closed-form Gradients for Fast Model Learning of Dynamic Systems. Henrik Krauß, Tim-Lukas Habich, Max Bartholdt, Thomas Seel, Moritz Schappler
We introduce the Domain-Decoupled Physics-Informed Neural Network (DD-PINN) to overcome the current limitations of PINNs for state-space models of large and complex dynamical systems. The time domain is decoupled from the neural network to construct an approach function that allows the computation of closed-form gradients. This approach significantly shortens training times, especially for large dynamic systems.
We would like to thank the entire team, especially Henrik Krauß, for their dedicated work on this project. Your collaboration and hard work has played a crucial role in this success.
You can find more information here: icinco.scitevents.org