Data-driven multiscale modeling for correcting dynamical systems

This repository stores the code associated with our paper also available on arXiv and which appeared at an earlier stage at the ICLR 2023 workshop on Tackling Climate Change with Machine Learning where it won "Best ML Innovation".

The code in this repository can be used to recreate our experiments or to modify our approach to work with additional systems. During our own work we made use of the Apptainer (or alternatively Singularity) container system, and our definition file closure.def can be used to build a container including JAX and required dependencies.

$ apptainer build closure.sif closure.def

This produces a container image closure.sif which can be used to run our software. For manual environment setup, dependencies are listed in requirements.txt and can be installed using pip.

Citing

If you make use of this software, please cite the associated paper:

@article{multiscaleclosure25,
  author={Karl Otness and Laure Zanna and Joan Bruna},
  title={Data-driven multiscale modeling for correcting dynamical systems},
  journal={Machine Learning: Science and Technology},
  year={2025},
  doi={10.1088/2632-2153/ae1a36}
}

License

The software in this repository is made available under the terms of the MIT license. See LICENSE.txt for details.