Table 1.

Summary of the differences in generative model across Applications 1–3.

Application		${\mathbf{\text{Q}}}_1^{(2)}$	${\mathbf{\text{Q}}}_2^{(2)}$	${\mathbf{\text{Q}}}_3^{(2)}$	${\mathbf{\text{Q}}}_4^{(2)}$	${\mathbf{\text{Q}}}_{11}^{(1)}$	${\mathbf{\text{Q}}}_{21}^{(1)}$	${\eta }_1^{(2)},{\Omega }_1^{(2)}$	${\eta }_1^{(1)},{\Omega }_1^{(1)}$	${\eta }_2^{(1)},{\Omega }_2^{(1)}$
1	EEG	Ip	–	–	–	In	–	−4, 16	−4, 16	–
	MEG	Ip	–	–	–	–	In	−4, 16	–	−4, 16
	EEGu	Ip	–	–	–	In	In	−4, 16	−4, 16	+4, 1/16
	MEGu	Ip	–	–	–	In	In	−4, 16	+4, 1/16	−4, 16
	E + MEG	Ip	–	–	–	In	In	−4, 16	−4, 16	−4, 16
2	E + MEG + fMRI	Ip	G1	G2	G3	In	In	−4, 16	−4, 16	−4, 16
3	Group-optimized	Ip	G1	G2	G3	In	In	−4, 16	−4, 16	−4, 16
	E + MEG + fMRI	$\widetilde{\mathbf{\text{Q}}}$

I_p, I_n represent p × p, n × n identity matrices, given p sources and n sensors. G_h represents a p × p matrix whose elements are zero except for those on the leading diagonal that correspond to vertices within the hth fMRI cluster, which are set to one (see Application 2)$\widetilde{\mathbf{\text{Q}}}$. represents a p × p matrix that is a linear combination of the source covariance components ${\mathbf{\text{Q}}}_h^{(2)}$, the linear weightings of which are the hyperparameter estimates from group-optimization (see Application 3). For remaining symbols, see main text.