Fig 1.
An illustration of unsupervised multi-source domain adaptation (UMDA) problems.
(a) illustrates UMDA problem with observable source data, and (b) illustrates data-free UMDA problem with no observable source data. It is challenging to reduce the distribution discrepancy between source and target domains in (b) since there are no accessible source data.
Table 1.
Comparison of DEMS and other methods.
Table 2.
Table of frequently-used symbols.
Fig 2.
DEMS shows the best classification accuracy for five target domains; each percentage indicates the accuracy increase compared to the second-best one for each target domain.
Table 3.
Comparison of different latent space transformation methods for unsupervised multi-source domain adaptation (UMDA).
Fig 3.
Overall architecture of DEMS.
Table 4.
Summary of datasets.
Fig 4.
Sample images (10 classes).
Table 5.
Classification accuracy of DEMS and baselines.
Table 6.
Ablation study on MNIST-M target dataset.
Fig 5.
Visualization of image adaptation from MNIST-M to other source domains.
Fig (a) enumerates target samples for Figs (b), (c), (d), and (e). The target samples are adapted by adaptation networks which are trained with different losses. For DEMS (Fig (b)), the adaptation gradually focuses on the close source domains (MNIST, SVHN, and SynDigits), resulting in performance enhancement. For (Fig (c)), some classes (digits 3, 7, and 9) are failed to be adapted to source domains. For
and
(Figs (d) and (e)), the adaptations are not trained at all.
Fig 6.
Sensitivity of accuracy to the hyperparameters ϵ (Eq 10) and λ (Eqs 2 and 6).