One problem about dsbn HOT 3 CLOSED

BN indicates that the BN layer is shared by both the source and target datasets. On the other hand, DSBN means that there are separate BN layers for source and target, and each layer is learned separately.
In the case of multiple sources, it means that each source has a separate BN.
For example, in the case of "separate", if there are 2 sources and 1 target, there are 3 BN layers.
In the case of "merged", if there are 2 sources and 1 target, 2 sources are considered as one source and there are 2 BN layers in total.

from dsbn.

Thank you for your reply. In the separate case of your example, the number of branch is 3, same for BN and DSBN. Why the result of BN and DSBN is different for the Table3?

from dsbn.

No. In the separate case of your example, the number of BN branches is not the same for BN and DSBN.
For example, Table3 in the paper,
A->W :
BN: one BN
DSBN: two BN (1 source, 1 target)
(A+D)->W: [two sources are merged into one and considered as a new dataset]
BN: one BN
DSBN: two BN (1 source, 1 target)
A, D -> W: [two sources and one target are inferenced with different batch construction]
BN: one BN
DSBN: three BN (2 sources, 1 target)

from dsbn.