Unsupervised Domain Adaptation Using Exemplar-SVMs with Adaptation Regularization

Table 1

Classification accuracies of different methods for different tasks of domain adaptation. We conduct the experiments on conventional transfer learning methods. Comparing with traditional methods, DAESVMs gain a big improvement in the prediction accuracy. And they also improve confronted with the approach of LRESVM which is proposed recently [average ± standard error of accuracy (%)].

Task	SVM	KMM	TCA	TJM	SA	GFK	LSSA	LRESVM	DAESVMs
	45.4	42.2	45.3	56.9	51.8	49.6	54.8	79.8
	50.7	42.7	60.3	56.4	56.4	55.7	57.3	74.9
	47.4	42.4	61.3	51.0	54.7	56.9	56.7	75.4
	50.7	48.3	54.7	58.6	57.1	51.2	58.4	77.2
	53.2	53.5	56.4	57.4	59.0	57.1	59.1	87.1
	44.2	45.8	50.4	58.8	62.7	57.1	58.1	74.1
	40.8	42.2	53.8	46.1	58.9	59.2	58.4	80.4
	48.3	41.6	43.9	49.6	54.3	59.4	57.7	79.0
	67.8	72.9	82.4	82.0	83.4	80.2	87.1	91.0
	42.4	41.9	53.0	50.8	57.0	66.2	59.7	74.3
	41.2	39.0	53.7	54.8	34.7	52.4	54.2	70.6
	80.2	82.0	87.9	83.4	78.9	81.2	87.2	89.2
Average	51.0	49.5	58.6	58.8	59.1	60.5	62.4	79.4