Algorithm 1: pseudo code of AGGD
1	Tensor ${\mathit{W}}_{\mathbf{0}}$, Tensor ${\mathit{L}}_{\mathbf{0}}$
2	Parameters: ${\mathit{W}}_{\mathbf{0}}$: Weight Parameters of Neural Network Nodes in the Previous Training Epoch; ${\mathit{L}}_{\mathbf{0}}$: Loss Values of the Neural Network for the Previous n Training Epochs.
3	Operation: The algorithm is based on the Pytorch framework, please refer to Pytorch [47] for related torch operations.
4	Output: W: Weight Parameters of Neural Network Nodes in this Training Epoch.
5	def AGGD(self, ${\mathbf{W}}_{\mathbf{0}}$,${\mathit{L}}_{\mathbf{0}}$):
6	# Initialization
7	W = ${\mathit{W}}_{\mathbf{0}}$
8	optimizer = torch.optim.Adam(model.parameters(), lr = 0.0004)
9	yPred = model(x)  #Predict output using previous generation weights
10	l = torch.nn.MSE(yPred, yReal)
11
	$yPredVariance=\frac{1}{mn}{\sum }_{i=1}^n{{\sum }_{j=1}^m\left(y_{i}Pred-Mean\left(\mathit{yPred}\right)\right)}^2,\left(yPre{d}_i\subset \mathit{yPred}\right)$
12	L = ${\mathit{L}}_{\mathbf{0}}$.append(l)
13	$Lvariance=\frac{1}{n}{\sum }_i^n{\left(l_i-Mean\left(\mathit{L}\right)\right)}^2,\left(l_i\in \mathit{L}\right)$
14	# Update network weights
15	$flag=count\ge 150\&l>0.006\&Lvariance<0.1\&yPredVariance<0.2$
16	if flag:
17	$\mathit{W}=torch.random.initWeight\left({\mathit{W}}_{\mathbf{0}}\right)while\mathit{W}={\mathit{W}}_{\mathbf{0}}$
18	count = 0
19	L = []
20	else:
21	optimizer.zero_grad()
22	loss.backward()
23	count += 1
24	valid()