Algorithm 1 DANN Training

Inputs: initial source domain $\mathcal{S}$, target domain $\mathcal{T}$ hyperparameters for FE, C, and D networks $h_{FE},h_C,h_D$ number of epochs for three steps of the training process $n_1,n_2,n_3$ number of training cycles n     Algorithm:     ${\mathcal{S}}_{labeled},{\mathcal{S}}_{unlabeled}\leftarrow$ separate data from $\mathcal{S}$ for labeled and unlabeled training     $X_{FEC}\leftarrow$ extract inputs from ${\mathcal{S}}_{labeled}$     $Y_{FEC}\leftarrow$ extract outputs from ${\mathcal{S}}_{labeled}$     $X_{FED}\leftarrow$ combine inputs from $\mathcal{T}$ and ${\mathcal{S}}_{unlabeled}$     $Y_{FED}\leftarrow$ create domain labels from $\mathcal{T}$ and ${\mathcal{S}}_{unlabeled}$     $FE,C,D\leftarrow$ initialize networks with $h_{FE},h_C,h_D$     for i < n do     $FEC\leftarrow FE+C$     train $FEC$ by $(X_{FEC},Y_{FEC})$ for $n_1$ epochs     $FE\leftarrow$ FE part of $FEC$     $C\leftarrow$ C part of $FEC$     $FED\leftarrow FE+D$     train $FED$ by $(X_{FED},Y_{FED}$ for $n_2$ epochs with $FE$ part frozen     $D\leftarrow$ D part of $FED$     $Y_{FE{D}_{inv}}\leftarrow$ invert $Y_{FED}$     train $FED$ by $(X_{FED},Y_{FE{D}_{inv}}$ for $n_3$ epochs with D part frozen     $FE\leftarrow$ FE part of $FED$     end for