Research map / Probabilistic and scientific modeling / Dynamical systems
Hamiltonian Generative Networks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Learn dynamics from image sequences using a latent Hamiltonian model rather than PixelHNN | HGN · Mass-spring, Pendulum, Two-body, Three-body | average pixel MSE: Mass-spring: 104.37 \pm 52.51 -> 6.23 \pm 2.03; Pendulum: 64.62 \pm 0.87 -> 11.72 \pm 4.14; Two-body: 78.9 \pm 1.93 -> 1 | Helped |
| Use Euler instead of leapfrog integration (ablation) | HGN · Mass-spring, Pendulum, Two-body, Three-body | average pixel MSE: Mass-spring: 6.23 \pm 2.03 -> 6.2 \pm 2.69; Pendulum: 11.72 \pm 4.14 -> 10.93 \pm 4.32; Two-body: 16.47 \pm 7.15 -> 15.0 | Mixed |
| Remove posterior sampling (ablation) | HGN · Mass-spring, Pendulum, Two-body, Three-body | average pixel MSE: Mass-spring: 6.23 \pm 2.03 -> 3.07 \pm 1.06; Pendulum: 11.72 \pm 4.14 -> 10.68 \pm 3.19; Two-body: 16.47 \pm 7.15 -> 14. | Helped |
| Remove the network mapping posterior samples to the initial state (ablation) | HGN · Mass-spring, Pendulum, Two-body, Three-body | average pixel MSE: Mass-spring: 6.23 \pm 2.03 -> 7.04 \pm 2.55; Pendulum: 11.72 \pm 4.14 -> 13.98 \pm 4.94; Two-body: 16.47 \pm 7.15 -> 16. | Hurt |
Closest papers by meaning
- Hamiltonian Score Matching and Generative Flows (NeurIPS 2024 · similarity 0.91)
- Learning Hamiltonian Dynamics at Scale: A Differential-Geometric Approach (ICML 2026 · similarity 0.89)
- SyMetric: Measuring the Quality of Learnt Hamiltonian Dynamics Inferred from Vision (NeurIPS 2021 · similarity 0.89)
- Frequency-Structured Hamiltonian Neural Network for Multi-Timescale Dynamics (NeurIPS 2026 · similarity 0.88)
- Learning long range dependencies through time reversal symmetry breaking (NeurIPS 2025 · similarity 0.87)
- Towards Cross Domain Generalization of Hamiltonian Representation via Meta Learning (ICLR 2024 · similarity 0.87)
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.