Skip to content

Latest commit

 

History

History
77 lines (50 loc) · 2.47 KB

README.md

File metadata and controls

77 lines (50 loc) · 2.47 KB
Raw

RegNeuralDE

Official Implementation of the ICML 2021 Paper Opening the Blackbox: Accelerating Neural Differential Equations by Regularizing Internal Solver Heuristics

USAGE

Experiments provided here were developed and tested on Julia v1.5.3. All other package versions are automatically enforced. To install do the following in Julia REPL:

] dev https://github.com/avik-pal/RegNeuralDE.jl

The code will be downloaded in the JULIA_PKG_DEVDIR directory.

CITATION

If you found this codebase useful in your research, please consider citing

@InProceedings{pmlr-v139-pal21a,
  title = 	 {Opening the Blackbox: Accelerating Neural Differential Equations by Regularizing Internal Solver Heuristics},
  author =       {Pal, Avik and Ma, Yingbo and Shah, Viral and Rackauckas, Christopher V},
  booktitle = 	 {Proceedings of the 38th International Conference on Machine Learning},
  pages = 	 {8325--8335},
  year = 	 {2021},
  editor = 	 {Meila, Marina and Zhang, Tong},
  volume = 	 {139},
  series = 	 {Proceedings of Machine Learning Research},
  month = 	 {18--24 Jul},
  publisher =    {PMLR},
  pdf = 	 {http://proceedings.mlr.press/v139/pal21a/pal21a.pdf},
  url = 	 {http://proceedings.mlr.press/v139/pal21a.html},
}

DATASETS

  • Preprocessed Physionet Data can be downloaded from here. Place the downloaded file in data/physionet.bson.

EXPERIMENTS

Important Parameters of the Experiments are controlled using the yml files in experiments/configs.

SUPERVISED CLASSIFICATION USING NEURAL ODE

Parameters controlled by experiments/configs/mnist_node.yml. To train a Vanilla/Regularized Neural ODE for MNIST classification:

$ julia --project=. experiments/mnist_node.jl

LATENT ODE FOR TIME SERIES INTERPOLATION

Parameters controlled by experiments/configs/latent_ode.yml. To train a Vanilla/Regularized Latent ODE with GRU Encoder for Physionet Time Series Interpolation

$ julia --project=. experiments/latent_ode.jl

TOY NEURAL SDE

To train a Vanilla and Regularized Neural SDE

$ julia --project=. experiments/sde_toy_problem.jl

SUPERVISED CLASSIFICATION USING NEURAL SDE

Parameters controlled by experiments/configs/mnist_nsde.yml. To train a Vanilla/Regularized Neural ODE for MNIST classification:

$ julia --project=. experiments/mnist_nsde.jl