Latent-Lagrangian Neural Networks for Reduced Order Modeling of Non-autonomous Nonlinear Dynamical Systems
The paper trains reduced-order dynamics in a learned physical latent space without using an ODE solver during training.
Agrawal and Thorin propose a latent Lagrangian framework for forced nonlinear dynamical systems. The model learns latent coordinates plus separate neural networks for kinetic and potential energy, using force supervision and virtual work to keep the latent dynamics physically consistent. The authors report that it captures nonlinear, non-convex dynamics and generalizes to unseen forces and initial conditions. ArXiv · AI/CL/LG's note
Agrawal and Thorin propose a latent Lagrangian framework for forced nonlinear dynamical systems. The model learns latent coordinates plus separate neural networks for kinetic and potential energy, using force supervision and virtual work to keep the latent dynamics physically consistent. The authors report that it captures nonlinear, non-convex dynamics and generalizes to unseen forces and initial conditions. ArXiv · AI/CL/LG's note
score 4