Codes for the TMLR paper Effective Latent Differential Equation Models via Attention and Multiple Shooting

Video Presentation

Instructions for running the experiments

1. Download and install Julia 1.8.5:
   https://julialang.org/downloads/oldreleases/

2. Install all dependencies:
   julia Experiments/scripts/install_dependencies.jl

3. Generate train and test datasets:
   julia Experiments/scripts/datasets_generation.jl

4. Train models
   julia Experiments/scripts/Training/train_goku.jl
julia Experiments/scripts/Training/Baselines/LSTMs/train_lstm.jl
julia Experiments/scripts/Training/Baselines/LatentODE/train_latent_ode.jl

Notice that when ran on a personal computer, these trainings could take weeks. Alternatively, if you have access to a SLURM cluster, you can adapt the corresponing batch scripts to your setup and run the trainings there:

sbatch Experiments/scripts/Training/batch_job_goku.sh
sbatch Experiments/scripts/Training/Baselines/LSTMs/batch_job_lstm.sh
sbatch Experiments/scripts/Training/Baselines/LatentODE/batch_job_latent_ode.sh

5. Evaluate trained models on test data
   julia Experiments/scripts/Evaluation/evaluate.jl

6. Plot results and perform statistical tests
   julia Experiments/scripts/Evaluation/statistics_and_plots.jl

Data availability

Due to privacy limitations to share the fMRI dataset used in the study, this repository includes only the experiments conducted on the simulated Stuart-Landau oscillators.

Citation

If you find this codebase useful, please consider citing:

@article{abrevaya2024effective,
  title = {Effective Latent Differential Equation Models via Attention and Multiple Shooting},
  author = {Germ{\'a}n Abrevaya and Mahta Ramezanian-Panahi and Jean-Christophe Gagnon-Audet and Pablo Polosecki and Irina Rish and Silvina Ponce Dawson and Guillermo Cecchi and Guillaume Dumas},
  journal = {Transactions on Machine Learning Research},
  issn = {2835-8856},
  year = {2024},
  url = {https://openreview.net/forum?id=uxNfN2PU1W}
}