Skip to main content
. Author manuscript; available in PMC: 2012 Mar 1.
Published in final edited form as: Neuroimage. 2010 Dec 2;55(1):113–132. doi: 10.1016/j.neuroimage.2010.11.037

Table 1.

Distributions for the nodes of the graphical model obtained using Eq. (47). Derivations are shown in Appendix D, where matrices Q, P1, P2, and W(u) and the cov(·) operator are defined. The matrix R(k, q) is a kq × kq permutation matrix with the property R(k, q)vec(ZT)=vec(Z) (the matrix R(t1, q) is defined analogously).

Functional form Parameters
q(S) = 𝒩(vec(S)|vec(〈S〉),It1⊗ΣS) S〉 = ΣS(〈α1LTM + 〈1〉Diag(〈w〉)CX〉)
ΣS = (〈α1LTL + 〈1In)−1
q(H) = 𝒩(vec(H)|vec(〈H〉),In⊗ΣH) H〉 = ΣH(〈α2BTY + 〈2〉〈ZTCTDiag (〈w〉))
ΣH = (〈α2BTB + 〈2Ik)−1
q(X)= 𝒩(vec(X)|vec(〈X〉), Σx) vec(〈X〉) = 〈1〉 Σx(It1CTDiag(〈w〉))vec(〈S〉)
ΣX=(1(It1Q)+R(t1,q)T(Diag(β1)T1TT1)R(t1,q))1
q(Z) = 𝒩(vec(Z) | vec(〈Z〉), ΣZ) vec(〈Z〉) = 〈2〉 ΣZ(IkCTDiag(〈w〉))vec(〈HT)
ΣZ=(2(IkQ)+R(k,q)T(Diag(β2)T2TT2)R(k,q))1
q(w) = 𝒩(w | 〈w〉, Σw) w〉 = Σwdiag(〈1〉 〈S〉 〈XTCT + 〈2〉〈HTZTCT)
Σw = (〈1P1 + 〈2P2 + 〈γW(u))−1
q(α1)= Γ(α1|aα1, bα1) aα1=mt12+aα10
bα1=12tr((MLS)T(MLS))+t12tr(ΣSLTL)+bα10
q(α2) = Γ(α2 | aα2, bα2) aα2=nt22+aα20
bα2=12tr((YBH)T(YBH))+n2tr(ΣHBTB)+bα20
q(1) = Γ(1|a1, b1) a1=t1n2
b1=12[tr(STS2STDiag(w)CX+XTQX)+t1tr(ΣS)+tr(ΣX(It1Q))]
q(2) = Γ(2|a1, b1) a2=kn2
b2=12[tr(HHT2HDiag(w)CZ+ZTQZ)+ntr(ΣH)+tr(ΣZ(IkQ))]
q((β1)i) = Γ((β1)i|(aβ1)i, (bβ1)i) (aβ1)=t12+aβ10
(bβ1)i=12Xi.T1TT1Xi.T+12tr(T1TT1cov((Xi.)T))+δ1
q((β2)i) = Γ((β2)i|(aβ2)i, (bβ2)i) (aβ2)i=k2+aβ20
(bβ2)i=12Zi.T2TT2Zi.T+12tr(T2TT2cov((Zi.)T))+δ2
q(δ1) = Γ(δ1|aδ1, bδ1) aδ1 = aβ10q
bδ1=Σi=1q(β1)i
q(δ2)= Γ(δ2|aδ2, bδ2) aδ2 = aβ20q
bδ2=Σi=1q(β2)i
q(γ) = Γ(γ|aγ, bγ) aγ = φn
bγ=Σi=1nui