Table 1. Properties of feature representation transfer methods.
DA type | Properties | ||||||
---|---|---|---|---|---|---|---|
Method | Unsup. | Semis. | Sup. | D ≥ 2 | Unpaired | dS ≠ dT | Nonlinear |
PCA [24] | √ | √ | |||||
KPCA [25] | √ | √ | √ | ||||
TCA [10] | √ | √ | √ | ||||
SSTCA [10] | √ | √ | √ | ||||
JDA [26] | √ | √ | √ | ||||
CCA [22] | √ | √ | √ | ||||
kCCA [9] | √ | √ | √ | √ | |||
MA [20] | p | √ | |||||
GM [27] | √ | √ | |||||
OT-lab [15] | √ | √ | √ | ||||
SGF [12] | √ | √ | √ | √ | √ | ||
GFK [13] | √ | √ | √ | √ | √ | ||
MMDT [18] | √ | √ | |||||
SSMA [23] | √ | √ | √ | √ | |||
KEMA | √ | √ | √ | √ | √ |
D: number of domains.
dS, dT: number of features in source and target.
: semilabels predicted by a classifier.
p: known corresponding samples, but no labels.