Research map / Probabilistic and scientific modeling / Differential equations and PDEs
Deep Energy-based Modeling of Discrete-Time Physics
Takashi Matsubara, Ai Ishikawa, Takaharu Yaguchi
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Train with discrete gradient (DGNet) instead of HNN++ with adaptive Dormand-Prince | energy-based PDE model · Cahn-Hilliard | time-derivative MSE (x1e0): 33.03 -> 7.14 | Improved |
| Train with discrete gradient (DGNet) instead of HNN++ with adaptive DP | energy-based PDE model · KdV | energy MSE (x1e-3): 3.01 -> 1.60 | Improved |
| Use implicit discrete-gradient prediction instead of explicit RK2 prediction (ablation) | DGNet · Cahn-Hilliard | energy MSE (x1e-6): >10000 -> 0.34 | Improved |
| Use implicit discrete-gradient prediction instead of explicit RK2 prediction (ablation) | DGNet · Mass-Spring | energy MSE (x1e-3): 61.26 -> 0.62 | Improved |
| Use implicit scheme instead of ada. DP at prediction (ablation) | DGNet · KdV | energy MSE (x1e-3): 1.60 -> 1.60 | No clear effect |
| Use DGNet instead of HNN | energy-based ODE model · Pendulum | energy MSE (x1e-3): 16.55 -> 10.79 | Improved |
| Use DGNet instead of SRNN (separable-Hamiltonian design) | energy-based ODE model · 2-Body | energy MSE (x1e-6): 40.37 -> 81.03 | Hurt |
| Use DGNet instead of HNN | energy-based ODE model · Real Pendulum | energy MSE (x1e-3): 2.86 -> 0.50 | Improved |
| Use DGNet instead of HNN | energy-based ODE model · Mass-Spring | time-derivative MSE (x1e-3): 38.22 -> 38.50 | No clear effect |
| Use higher-order adaptive integrator (ada. DP) instead of RK2 for HNN++ training | HNN++ · Cahn-Hilliard | time-derivative MSE (x1e0): 344.23 -> 33.03 | Improved |
Closest papers by meaning
- Hamiltonian Neural PDE Solvers through Functional Approximation (NeurIPS 2025 · similarity 0.85)
- Physics-Informed Regularization for Domain-Agnostic Dynamical System Modeling (NeurIPS 2024 · similarity 0.85)
- Physics-Informed DeepONets for drift-diffusion on metric graphs: simulation and parameter identification (ICML 2025 · similarity 0.85)
- Riemann Tensor Neural Networks: Learning Conservative Systems with Physics-Constrained Networks (ICML 2025 · similarity 0.85)
- Harnessing the Power of Neural Operators with Automatically Encoded Conservation Laws (ICML 2024 · similarity 0.84)
- Learning Deep Dissipative Dynamics (AAAI 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
