Algorithm 2 Our Proposed Model

Input: source domain data $X_s$, target domain data $X_t$, labels $y_S$ of source domain data, threshold $\delta$, $\alpha$ and $\beta$

Output:$y_T$ as labels of target domain data

1: BEGIN

2: Initialize ${\mathcal{D}}_t^{\left(c\right)}$ as Null

3: while not converged do

4:   (1) $W$ $\leftarrow$ Distribution adaptation $\left({\mathcal{D}}_t^{\left(c\right)}\dots {\widehat{y}}_T\right)$ in Equation (14) and let $Z_s=W^T{X}_s$ and $Z_t=W^T{X}_t$

5:    (2) Assign ${\widehat{y}}_T$ using classifiers trained by $Z_s$

6:    (3) Obtain $Bnd(X)$ in Algorithm 1

7:     (4) $\left({\mathcal{D}}_t^{\left(c\right)}\dots {\widehat{y}}_T\right)$ $\leftarrow$ execute label propagation algorithm

8: End while

9:$y_T$$\leftarrow$${\widehat{y}}_T$

10: END BEGIN