Code for paper on "Anonymizing Personalization in Virtual Brain Twins" accepted for ICANN 2025.
Broadly the workflow is
- download/format data
- train the cross-coder
- run simulations
- for fixed connectomes: "subject-level"
- for connectomes sampled from cross-coder based prior: "cohort-level"
- run sbi
- compare diagnostics
At the moment, the dynamical regimes and data features used to demo this are minimal and could be expanded.
pip install matplotlib tqdm vbjax sbi typed-argparse joblib
If you have Git LFS installed, you can just clone the repo and the cross coder will already be trained.