A second-order inspired neural ordinary differential equation (neural ODE) model of double-pendulum dynamics [1].
The implementation follows the double-pendulum benchmark from the original publication, with two differences.
First, a 70/30 split is used for training and validation data instead of the 50/50 split used in the original study, as the latter led to non-correlation between training and validation loss.
Second, the original publication scales the time span by the maximum training time. This scaling is implemented directly in the PEtab measurement tables. For example, the final training time point is 1.0.
Both training and validation data are provided. The corresponding PEtab-SciML problems can be built using the respective YAML files.
The provided nominal values do not correspond to an optimal fit, so lower loss values can be achieved when training the model.
- Ko, Joon-Hyuk, et al. "Homotopy-based training of neuralodes for accurate dynamics discovery." Advances in Neural Information Processing Systems 36 (2023): 64725-64752.