model parameters are estimated by maximising the posterior

2006], giving the iterative variance update formula shown below,

ߪ௠^ଶൌ

∑ߴ௜,௠ݖ௜

ଶ

௜

൅2ܾ௠

∑ߴ௜,௠

௜

൅2ܽ௠൅2

(6.51)

௠ ^{is defined as below,}

ߴ௜,௠ൌ^{ݓ௠࣡ሺࢠ࢏|ߤ௠, ߪ௠ଶሻ}

݂ሺࢠ࢏|ߠሻ

(6.52)

erative update rule for the mixing coefficients can be written as,

ݓ௠ൌ¹

݊^෍ߴ௜,௠

௡

௜ୀଵ

(6.53)

erative update rule for each centre is, where ߤ௠଴ൌ0,

ߤ௠ ൌ^{ߚ௠∑ߴ௜,௠ݖ௜}

௜

൅2߭௠ߤ௠଴

ߚ௠∑ߴ௜,௠

௜

൅2߭௠

(6.54)

(a) (b)

he demo of how model parameters converge in a DSG model. (a) The standard

values for both the null density (the lower cluster of the curves) and the

density (the upper cluster of the curves). (b) The mixing coefficient values for

ll density (the lower cluster of the curves) and the alternative density (the upper

he curves).