Table 1.

Tested and optimal hyperparameter values.

Parameter	Tested	Optimal
Number of latent sources	16,32,64	32
Temporal filter length	3,5,7,9,11	7
Learning rate	1⋅10−3, 3⋅10−4, 1⋅10−4	3⋅10−4
$l_1$ -penalty	1⋅10−3, 3⋅10−4, 1⋅10−4	3⋅10−4
Pooling	max	max
Pooling factor	2,3,5	2
Drop-out coefficient	0.25, 0.50, 0.75, 0.90	0.50
Input layer link function	identity, ReLU, ELU, tanh	identity
Hidden layer link function	identity, ReLU, ELU, tanh	ReLU
Output nonlinearity	sigmoid, softmax	softmax
Number of dense hidden layers	1,2	1