1: Initialize the encoder parameters ${\Theta }_{E_r}$ , weights and biases of n adaption layers: $A_{{\Theta }_{A_i}}=\left\{w_{A_i},b_{A_i}\right\}\left(i=1,\cdots ,n\right)$ . Freeze all parameters of $D_{{\Theta }_{D_1}},\cdots ,D_{{\Theta }_{D_n}}$ .

2: for each epoch do

3: Sample $v_i\in d_i\left(i=1,\cdots ,n\right)$

4: Traverse $\theta \in \left\{{\Theta }_{E_r},{\Theta }_{A_1},\cdots ,{\Theta }_{A_i},\cdots ,{\Theta }_{A_n}\right\}$

5: Apply Backpropagation to update $\theta$ for all adaption layers and encoder.

$\theta \leftarrow \theta -\eta {\nabla }_{\theta }\left\{{\displaystyle \sum _{i=1}^n{\mathbb{E}}_{v_i}\mathcal{L}\left(v_i,D_{{\Theta }_{D_i}}\left[A_{{\Theta }_{A_i}}\left(E_{{\Theta }_{E_r}}\left(v_i\right)\right)\right]\right)}\right\}$ (8)

where $v_i\in \mathcal{V}\left\{v_1,\cdots ,v_i,\cdots ,v_n\right\}\left(i=1,\cdots ,n\right)$