ResKoopNet

---

Experiments 1: Pendulum

I have provided two versions of Pendulum system using ResKoopNet and a comparison test with other 4 methods:

1. The pendulum.m file is used to generate pendulum data,
2. TensorFlow Version: pendulum_reskoopnet_tf.ipynb,
3. PyTorch Version: pendulum_reskoopnet_torch.ipynb,
4. Comparison test: pendulum_edmd_reskoopnet_hankeldmd_edmddl.ipynb,
5. Algorithm of Hankel-DMD: hankel_dmd_utils.py.
6. Some dataset for Pendulum example: here.

Experiments 2: Turbulence

We have computed Koopman modes using ResKoopNet and Hankel-DMD in these experiments:

1. Experiment on turbulence using ResKoopNet: turbulence_reskoopnet.ipynb,
2. Plotting Koopman modes computed from ResKoopNet method: turbulence_reskoopnet_plot.ipynb,
3. Experiment on turbulence using ResKoopNet with plotting: turbulence_hankeldmd.ipynb.

Experiments 3: Neural dynamics identification in mice visual cortex

We have tested on the open dataset on mice from the competition ``Sensorium 2023" via comparing ResKoopNet and several other classical methods in this experiment.

Solvers

1. The solver_edmd_tf.py is based on a modified version of the EDMD-DL solver. You can find more details here.
2. solver_resdmd_tf.py and solver_resdmd_torch.py are also modified based on the EDMD-DL solver.

Plotting Tools

1. koopman_pseudospec_qr - A plotting tool to visualize the pseudospectrum. This Python file is translated from here.

References

If you use ResKoopNet or the code in your research, please kindly cite the following paper:

@@inproceedings{xu2025reskoopnet,
  title     = {ResKoopNet: Learning Koopman Representations for Complex Dynamics with Spectral Residuals},
  author    = {Xu, Yuanchao and Shao, Kaidi and Logothetis, Nikos and Shen, Zhongwei},
  booktitle = {Proceedings of the 42nd International Conference on Machine Learning},
  year      = {2025},
  publisher = {PMLR},
  series    = {ICML '25},
  url       = {https://openreview.net/forum?id=Svk7jjhlSu}
}

---