Rethink GraphODE Generalization within Coupled Dynamical System
Authors: Guancheng Wan, Zijie Huang, Wanjia Zhao, Xiao Luo, Yizhou Sun, Wei Wang
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments across diverse dynamical systems demonstrate that ours outperforms state-of-the-art methods within both in-distribution and out-of-distribution. |
| Researcher Affiliation | Academia | 1University of California, Los Angeles 2Stanford University. Correspondence to: Guancheng Wan <EMAIL>. |
| Pseudocode | No | The paper describes the methodology using text and mathematical equations, but does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | The code is available at https://github. com/Guancheng Wan/GREAT. |
| Open Datasets | No | We evaluate GREAT using three coupled dynamical systems datasets: SPRING, CHARGED, and PENDULUM, which model the dynamics of physical systems with complex interdependencies. For evaluation, we adopt two metrics: RMSE (Root Mean Square Error) and MAPE (Mean Absolute Percentage Error). We assess GREAT under both in-distribution (ID) and out-of-distribution (OOD) settings, where the OOD setting modifies the test dataset s initial conditions (e.g., velocity, position) to test the model s ability to generalize to unseen scenarios. Further details are provided in Appendix A. The datasets are generated using a physics-based simulation framework, where the system parameters are sampled from the specified ranges. |
| Dataset Splits | Yes | The datasets are split into training, validation, and test sets, with additional out-of-distribution (OOD) test sets to evaluate model generalization. Further details are provided in Appendix A. ... The training, validation, and test sets are split as follows: 5000 samples for training, 1000 samples for validation, 1000 samples for testing, and 1000 samples for OOD testing. |
| Hardware Specification | Yes | The experiments are conducted using NVIDIA Ge Force RTX 3090 GPUs as the hardware platform, coupled with Intel(R) Xeon(R) Gold 6240 CPU @ 2.60GHz. |
| Software Dependencies | Yes | The deep learning framework employed was Pytorch, version 1.11.0, alongside CUDA version 11.3. |
| Experiment Setup | Yes | The hidden layer size was set to 32 for each dataset. For optimization, the Adam optimizer (Kingma & Ba, 2014) was chosen, with a learning rate of 1e 5 and a weight decay of 1e 3 during the training process. |