Algorithm 1 Training the VAE-WACGAN model
1:	${\theta }_E,{\theta }_G,{\theta }_D\leftarrow$ initialize network parameters
2:	repeat
3:	for $i=1$ $\to$ $n_{critic}$ do
4:	$(X_r,y_r)\leftarrow$ random mini-batch from dataset
5:	$z_r\leftarrow E(X_r,y_r)$
6:	$X_f\leftarrow G(z_r,y_r)$
7:	$z_p\leftarrow$$\mathrm{samples}\mathrm{from}\mathrm{prior}\mathcal{N}(0,I)$
8:	$X_p\leftarrow G(z_p,y_r)$
9:	// Update the parameters of the Discriminator
10:	${\mathcal{L}}_S\leftarrow -\mathbb{E}[{S(X}_r)]+\mathbb{E}[{S(X}_f)]+\mathbb{E}[{S(X}_p)]+\lambda \mathbb{E}[{(\parallel {\nabla }_{\hat{x}}S\left(\hat{x}\right){\parallel }_2-1)}^2]$
11:	${\mathcal{L}}_C\leftarrow -\left(\mathbb{E}[\log {P\left(C\right(X}_r{)=y}_r)]-\mathbb{E}[\log {P\left(C\right(X}_f{)=y}_r)]-\mathbb{E}[\log {P\left(C\right(X}_p{)=y}_r)]\right)$
12:	${\mathcal{L}}_D\leftarrow {\mathcal{L}}_S+{\mathcal{L}}_C$
13:	${\theta }_D\leftarrow \mathrm{Adam}({\theta }_D,{\nabla }_{{\theta }_D}({\mathcal{L}}_D\left)\right)$
14:	end for
15:	// Update the parameters of the Encoder and Decoder
16:	${\mathcal{L}}_{prior}\leftarrow D_{KL}\left(q\right(z_r|X_r,y_r)\parallel p\left(z_r\right|y_r\left)\right)$
17:	${\mathcal{L}}_{recon}\leftarrow -{\mathbb{E}}_{q\left(z_r\right|X_r,y_r)}\left[\log p\left(X_r\right|z_r,y_r)\right]$
18:	${\mathcal{L}}_E\leftarrow {\mathcal{L}}_{prior}+\mathrm{{\rm Y}}\cdot {\mathcal{L}}_{recon}$
19:	${\mathcal{L}}_G\leftarrow -\mathbb{E}[{S(X}_p)]-\mathbb{E}[\log {P\left(C\right(X}_p{)=y}_r)]+\mathrm{{\rm Y}}\cdot {\mathcal{L}}_{\mathrm{recon}}$
20:	${\theta }_E\leftarrow \mathrm{Adam}({\theta }_E,{\nabla }_{{\theta }_E}({\mathcal{L}}_E\left)\right)$
21:	${\theta }_G\leftarrow \mathrm{Adam}({\theta }_G,{\nabla }_{{\theta }_G}({\mathcal{L}}_G\left)\right)$
22:	until $D$ has converged to 0.5