Table 3.

Description of SMNN architectures.

SMNN	Hidden layer	Neuron type	Drop out	Hidden layer	Neuron type	Drop out	Hidden layer	Neuron type	Output layer	Neuron type
1	$N_{\mathit{stack}}\times 50\to 50$	ReLU							1	Sigmoid
2	$N_{\mathit{stack}}\times 50\to 100$	ReLU							1	Sigmoid
3	$N_{\mathit{stack}}\times 50\to 200$	ReLU							1	Sigmoid
4	$N_{\mathit{stack}}\times 50\to 50$	ReLU	20%	50	PReLU				1	Sigmoid
5	$N_{\mathit{stack}}\times 50\to 50$	ReLU	20%	100	PReLU				1	Sigmoid
6	$N_{\mathit{stack}}\times 50\to 50$	ReLU	20%	200	PReLU				1	Sigmoid
7	$N_{\mathit{stack}}\times 50\to 50$	ReLU	20%	50	PReLU	20%	50	PReLU	1	Sigmoid
8	$N_{\mathit{stack}}\times 50\to 50$	ReLU	20%	100	PReLU	20%	100	PReLU	1	Sigmoid
9	$N_{\mathit{stack}}\times 50\to 50$	ReLU	20%	200	PReLU	20%	200	PReLU	1	Sigmoid

Numbers correspond to the number of neurons in each layer. For example, SMNN 1 consists of one hidden linear layer with 50 ReLU neurons and a linear output layer with one sigmoid neuron. $N_{\mathit{stack}}\times 50$ denotes dimension of the input to the first hidden layer.