Table 8.

The flow of the CMMD algorithm.

Center Loss with MMD
Input: Training Data [${\mathrm{X}}_{Source}, {\mathrm{X}}_{Target}$] Label ${\mathrm{Y}}_{Source}$
Hyperparameter $\mathsf{\lambda },\mathsf{\alpha },\mathsf{\beta },$ learning rate ${\mu }^t$, epochs t
Output: Neural network parameter${\theta }_c$
1: while not converge do 2:  t$\leftarrow$t+1
3: compute total loss: ${\mathcal{L}}^t={\mathcal{L}}_s^t$ + ${\mathcal{L}}_C^t$ + ${\mathcal{L}}_m^t$ 4: compute gradient: $\frac{\partial {\mathcal{L}}^t}{\partial X_i^t}$ = $\frac{\partial {\mathcal{L}}_S^t}{\partial X_i^t}$ + $\mathsf{\lambda }·\frac{\partial {\mathcal{L}}_C^t}{\partial X_i^t}$ + $\mathsf{\beta }·\frac{\partial {\mathcal{L}}_m^t}{\partial X_i^t}$
5: update central points: $C_j^{t+1}=C_j^t-\alpha ·\mathsf{\Delta }{C}_j^{t+1}$ 6: update network parameter: ${\theta }_C^{t+1}={\theta }_C^t-{\mu }^t{\sum }_i^m\frac{\partial {\mathcal{L}}^t}{\partial X_i^t}·\frac{\partial X_i^t}{\partial {\theta }_C^t}$. 7: end while