This tool was developed at INS in Marseille.
Authors: Nina Baldy, Marmaduke Woodman, Viktor Jirsa, Meysam Hashemi

The aim is to provide inference services for Dynamical Causal Modeling of Event-Related Potentials (ERPs) measured with EEG/MEG, using SATO Probabilistic Programming Languages (PPLs):

Numpyro: https://num.pyro.ai/en/stable/

Blackjax: https://blackjax-devs.github.io/blackjax/

PyMC: https://www.pymc.io/welcome.html

Stan: https://mc-stan.org/

We have provided a taxonomy for model comparison tailored to algorithms: (1) adaptive Hamiltonian Monte Carlo, (2) automatic Laplace and (3) family of variational inference. We have provided solutions to address the deference by: 1) optimizing the hyperparameters, (2) leveraging initialization with prior information, (3) weighted stacking based on predictive accuracy.

Notebooks:

https://wiki.ebrains.eu/bin/view/Collabs/ebrains-task-3-3/Drive#notebooks/DCM_ERP_NumPyro

Tutorial:

https://wiki.ebrains.eu/bin/view/Collabs/ebrains-task-3-3/Drive#notebooks/EITN_tutorial 

@article{Baldy2024AutoDCM,
  title={Dynamic Causal Modeling in Probabilistic Programming Languages},
  author={Baldy, Nina and Woodman, Marmaduke and Jirsa, Viktor and Hashemi, Meysam},
  journal={bioRxiv},
  pages={2024--11},
  year={2024},
  publisher={Cold Spring Harbor Laboratory}
}