Stable time-delay systems

Requirements | Training | Contributing

This repository is the official implementation of Learning Stable Deep Dynamics Models for Partially Observed or Delayed Dynamical Systems.

Requirements

All requirements are summarized in the Project.toml file.

Training

This code implements stable Neural DDEs. To reproduce our results simply run the python scripts in src/experiments/:

• scr/experiments/train_ANODE_cos.jl
• scr/experiments/train_NDDE_cos.jl
• src/experiments/train_n_pendulum_LRF_alongtraj.jl
• src/experiments/train_n_pendulum_unstable.jl
• src/experiments/train_inverted_pendulum_LRF_alongtraj.jl

Contributing

If you would like to contribute to the project please reach out to Andreas Schlaginhaufen. If you found this library useful in your research, please consider citing.

@article{schlaginhaufen2021learning,
      title={Learning Stable Deep Dynamics Models for Partially Observed or Delayed Dynamical Systems},
      author={Andreas Schlaginhaufen, Philippe Wenk, Andreas Krause and Florian Dörfler},
      journal={Advances in Neural Information Processing Systems},
      year={2021},
}