Parameter exploration and inference
In this chapter you will fit whole-brain models to empirical observables using TVB-Optim’s gradient-based optimisation. Two complementary use cases are covered, each on a different modality and observable, but sharing the same TVB-Optim idioms (Parameter, .shape, Space, @cache).
1 Sessions
1.1 fMRI BOLD functional connectivity
Fit a Reduced Wong-Wang network on the Desikan-Killiany parcellation to match empirical resting-state FC. Workflow: grid exploration over (G, w), global gradient fit, then heterogeneous per-region w. Loss is the RMSE between simulated and empirical FC matrices.
1.2 MEG peak frequency gradient
Fit a Jansen-Rit network with delays to reproduce the resting-state alpha gradient (7 Hz in association areas to 11 Hz in visual cortex). Workflow: grid exploration over (a, b), then per-region (a, b) gradient fit against Cauchy target spectra.
See the workshop agenda.