Table 2.

Examined types of autoencoder neural networks.

Autoencoder architecture	Unit activation function	Regularization constraints on parameters	Number of latent layers	Units per latent layer	Tied weights
Baseline	Identity	None	1	15	No
Baseline + l1	Identity	l1 penalty	1	15	No
Baseline + l2	Identity	l2 penalty	1	15	No
Baseline + covariance	Identity	Covariance	1	15	No
Tied non-linear	Relu	None	1	15	Yes
3-layer non-linear	Relu	None	3	25-15-25	No
5-layer non-linear	Relu	None	5	25-20-15-20-25	No

Summary of the different types of artificial neural network classes that were explored to discover structural dependencies between social brain regions.