Research map / Probabilistic and scientific modeling / Dynamical systems
An Extensible Benchmark Suite for Learning to Simulate Physical Systems
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 |
|---|---|---|---|
| Increase training set size for neural-network function approximators | neural networks · Spring and Wave | MSE | No clear effect |
| Use noise injection during training | neural networks on Navier-Stokes · Navier-Stokes | stability | Helped |
| Use derivative prediction rather than step prediction | CNNs · Spring Mesh | MSE | Mixed |
| Evaluate with a higher-order integrator | learned derivative predictors · Navier-Stokes | MSE | Hurt |
| Evaluate with a higher-order integrator | learned derivative predictors · most systems | MSE | Helped |
| Use four obstacles rather than one | Navier-Stokes learning task · Navier-Stokes | MSE | Hurt |
| Use a larger network such as a u-net instead of KNN or MLP | Navier-Stokes learning task with four obstacles · Navier-Stokes, four obstacles | MSE | Helped |
| Emphasize early trajectory errors with per-step weighting | learned methods compared with KNN · Navier-Stokes, step prediction | MSE | Helped |
| Evaluate on out-of-distribution obstacle radii | all evaluated methods · Navier-Stokes | MSE | No clear effect |
| Increase numerical integration step size using the reported scaling | learned derivative prediction compared with numerical integrators · Spring | time ratio: 173.3; 1×; 198.6; 1×; 173.3; 1× | Mixed |
Closest papers by meaning
- EvoMesh: Adaptive Physical Simulation with Hierarchical Graph Evolutions (ICML 2025 · similarity 0.88)
- Space and time continuous physics simulation from partial observations (ICLR 2024 · similarity 0.87)
- Multiple Physics Pretraining for Spatiotemporal Surrogate Models (NeurIPS 2024 · similarity 0.87)
- Breaking the Discretization Barrier of Continuous Physics Simulation Learning (NeurIPS 2025 · similarity 0.86)
- The Well: a Large-Scale Collection of Diverse Physics Simulations for Machine Learning (NeurIPS 2024 · similarity 0.86)
- Generating Physical Dynamics under Priors (ICLR 2025 · similarity 0.86)
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.