Glossary#

This glossary defines CaliBrain-specific vocabulary together with general inverse-imaging and neuroimaging terms used throughout the documentation.

adaptive_joint_learning#: A noise-variance workflow mode in which no fixed sensor-noise variance is supplied to the inverse solver. Instead, a joint-learning solver estimates the noise level during fitting.
aggregated calibration#: Calibration performed on source summaries that have been reduced across time rather than on the full source-by-time posterior output.
baseline noise variance#: Noise-variance estimate obtained from the pre-stimulus segment of the simulated sensor data.
BMN#: Bayesian minimum norm. In CaliBrain, this refers to a minimum-norm-style Bayesian inverse solver that returns posterior means and covariance summaries under a common source-variance model.
BMN_joint#: A BMN variant that can learn a common sensor-noise variance jointly with the source hyperparameter.
calibration#: The process of comparing nominal uncertainty levels with empirical coverage and, when needed, learning a recalibration map.
calibration curve#: A curve showing empirical coverage as a function of nominal coverage.
coil type#: MNE/FIFF metadata describing the sensor hardware type, for example EEG electrodes or MEG magnetometers.
credible interval#: An interval derived from a posterior distribution. In CaliBrain, scalar credible intervals are used for fixed-orientation uncertainty evaluation.
credible set#: A posterior uncertainty region, such as an interval or ellipsoid, associated with a nominal coverage level.
EEG#: Electroencephalography. In CaliBrain, EEG can be treated in fixed orientation or in free orientation with three local source components.
EMD#: Earth mover’s distance. In CaliBrain, this is used as a source-space distributional metric for comparing estimated and true source activity.
empirical coverage#: The observed fraction of cases in which the true source value or vector lies inside the nominal posterior credible set.
forward model#: The mapping from source activity to sensor measurements, represented by the leadfield.
free orientation#: A source model in which each source location has multiple orientation components rather than a single fixed scalar coefficient.
full_cov#: CaliBrain’s name for the free-orientation uncertainty representation that uses local posterior covariance blocks to define multivariate ellipsoidal credible sets.
gamma_lambda_map_sflex#: An sFLEX Gamma-MAP variant that jointly learns a noise-related regularization parameter.
gamma_map_sflex#: A sparse Bayesian inverse solver in CaliBrain based on Gamma-MAP and an sFLEX basis construction.
head#: Informal workflow term for one subject-specific or geometry-specific simulation context used when pooling or splitting calibration data.
inverse problem#: The problem of recovering latent neural source activity from EEG/MEG sensor measurements.
isotonic regression#: A monotone regression method used in CaliBrain to recalibrate nominal coverage levels while preserving ordering.
leadfield#: The matrix or tensor that maps source amplitudes to sensor measurements.
marginal#: CaliBrain’s name for the free-orientation uncertainty representation that calibrates component-wise intervals using marginal variances only.
MEG#: Magnetoencephalography. In CaliBrain, MEG can be represented in fixed orientation or in a reduced free-orientation form with tangential components.
noise-variance strategy#: The rule used to provide or estimate the sensor-noise variance for source reconstruction. In CaliBrain, the main strategies are oracle, baseline, and adaptive_joint_learning.
nominal coverage#: The target coverage level attached to a credible interval or credible set, for example 0.9 for a nominal 90% credible set.
oracle noise variance#: The true sensor-noise variance computed from the injected simulation noise.
post_fixed#: A calibration workflow mode in which one recalibration map is fit at a reference condition and then reused across a sweep of evaluation conditions.
post_oracle#: A calibration workflow mode in which recalibration is fit and evaluated on matched train and evaluation conditions.
post_pooled#: A calibration workflow mode in which recalibration is fit on pooled training data and evaluated on a target condition.
post_pooled_mismatch#: A calibration workflow mode in which recalibration is fit on pooled but intentionally mismatched training conditions and evaluated on the target condition.
posterior covariance#: The covariance matrix returned by an inverse solver to quantify posterior uncertainty in source space or coefficient space.
posterior mean#: The mean of the posterior distribution returned by an inverse solver and used as the point estimate of source activity.
posterior summary#: The stored solver output used downstream in CaliBrain, typically including posterior mean, posterior covariance, and associated metadata.
precal#: A workflow mode that evaluates raw empirical coverage without fitting a recalibration map.
recalibration#: The post-hoc correction of nominal coverage levels using a fitted mapping such as isotonic regression.
reduced free-orientation MEG#: A free-orientation MEG representation with two tangential components per source location.
run manifest#: A tabular index of generated runs used to locate posterior summaries and their metadata in downstream workflow stages.
sensor noise#: Additive noise at the sensor level, simulated in CaliBrain before source reconstruction.
source activity#: Latent neural current amplitudes at source locations over time.
source space#: The discrete set of candidate source locations used for inverse source imaging.
source_estimator#: The high-level CaliBrain class that wraps an inverse solver and applies it to a leadfield and sensor data.
uncertainty representation#: The geometric object used for calibration, such as a scalar interval, component-wise interval family, or local ellipsoid.