Glossary#

Connectome-based neural mass model (CNMM)#: In this project, a neural mass model coupled by a structural connectivity matrix (sc), with optional distance-based delays (dist) and a measurement model.
Distance matrix#: Pairwise inter-regional distances (dist), used to compute integer delay indices in delayed coupling terms.
External input / stimulus#: The u or external input tensor supplied to forward simulation for resting-state (often zero) or evoked/stimulus-driven modeling.
Fitting paradigm#: A training workflow class (for example ModelFitting, FittingBatch, FittingFNGFPG) that defines simulation, optimization, and bookkeeping behavior.
FNG-FPG#: FittingFNGFPG (“Forward No Gradient, Forward Parallel Gradient”), a specialized approach for long simulations using a serial pass for initial conditions and a gradient-enabled blocked pass.
Lead field matrix#: A source-to-sensor projection matrix (lm) used to map simulated source activity to M/EEG channel space.
Neural mass model (NMM)#: A reduced biophysical model of population activity per region. In WhoBPyT, NMMs inherit from whobpyt.datatypes.AbstractNeuralModel.
Objective function / loss#: A differentiable criterion minimized during fitting, implemented via AbstractLoss and concrete classes such as CostsTS, CostsFC, and CostsMean.
Output variables#: Simulated observables exposed by a model for fitting. For the current Jansen-Rit implementation, output_names = ["eeg"].
Parameter object#: The whobpyt.datatypes.Parameter class storing a value, optional prior mean/precision, and flags controlling whether values and/or priors are trainable.
Prior loss#: A regularization term from AbstractLoss.prior_loss that penalizes deviation from parameter priors when hyperparameters are fit.
State variables#: Dynamical variables integrated by the model at each step. For JansenRitModel these are E, Ev, I, Iv, P, and Pv.
Structural connectivity (SC)#: The inter-regional coupling matrix used for long-range interactions (for example sc in JansenRitModel), often derived from diffusion MRI tractography.
Timeseries#: The whobpyt.datatypes.Timeseries container for empirical or simulated data, typically shaped as num_regions x ts_length.
Training statistics#: The TrainingStats record of losses, tracked parameters, and selected matrices (for example fitted connectivity/lead-field terms) across optimization.
Whole-brain model (WBM)#: A dynamical systems model that simulates activity across many brain regions at once, using region-to-region coupling and neural state equations.
Windowed fitting#: A training pattern where longer recordings are split into windows (Timeseries.windowedTensor), and gradients are backpropagated window-by-window.