Algorithm 1 Training Algorithm of SAD-SNN
Input: Dataset $\mathcal{D}$. Model: Teacher ANN model ${\mathcal{M}}_{\mathrm{ANN}}(x;W_{\mathrm{ANN}})$ with pretrained weights $W_{\mathrm{ANN}}$; Student SNN model ${\mathcal{M}}_{\mathrm{SNN}}(x;W_{\mathrm{SNN}})$ with initial weights $W_{\mathrm{SNN}}$. Parameter: Time steps T; Layer alignment set $\mathcal{I}$; Distillation loss weight $\alpha$; Learning rate $\eta$; Training epoch E. Output: Trained student SNN model ${\mathcal{M}}_{\mathrm{SNN}}(x;W_{\mathrm{SNN}})$.  1: for  $epoch=1$  to  E  do  2:     for x, y in $\mathcal{D}$ do
3:         ${\mathbf{A}}_{\mathcal{T}}={\mathcal{M}}_{\mathrm{ANN}}\left(x\right)$;	▹ forward propagation
4:         ${\mathbf{A}}_{\mathcal{S}},\widehat{y}={\mathcal{M}}_{\mathrm{SNN}}\left(x\right)$;
5:         ${\mathbf{A}}_{{\mathcal{S}}^{\prime }}={\mathrm{AvgPool}}_T\left({\mathbf{A}}_{\mathcal{S}}\right)$;	▹ calculate spatial-activation maps
6:         calculate ${\mathbf{S}}_{\mathcal{T}}$ and ${\mathbf{S}}_{{\mathcal{S}}^{\prime }}$ based on ${\mathbf{A}}_{\mathcal{T}}$ and ${\mathbf{A}}_{{\mathcal{S}}^{\prime }}$ using (4);
7:         calculate ${\mathcal{L}}_{distill\_SAD}$ based on ${\mathbf{S}}_{\mathcal{T}}$ and ${\mathbf{S}}_{{\mathcal{S}}^{\prime }}$
8:                               using layer pairs $(l,l^{\prime })\in \mathcal{I}$ via (5);	▹ calculate ${\mathcal{L}}_{distill\_SAD}$
9:         calculate task-specific loss based on y and $\widehat{y}$;	▹ calculate ${\mathcal{L}}_{task}$
10:          calculate ${\mathcal{L}}_{total}$ based on ${\mathcal{L}}_{task}$ and ${\mathcal{L}}_{distill\_SAD}$ using (6);	▹ calculate ${\mathcal{L}}_{total}$
11:          calculate the gradients ${\displaystyle \frac{\partial {\mathcal{L}}_{\mathrm{total}}}{\partial W_{SNN}}}$;	▹ backward propagation
12:          update $W_{SNN}$;	▹ update weights
13:      end for
14:  end for