Table 3.

Hyper-parameters of the deep-learning-based models found for $T=64$ on the OPPORTUNITY dataset. The parameters values are given assuming that the input of the models is of size $T\times S$. Convolutional kernels and pooling sizes are, respectively, given as $(f_T^{(k)},f_S^{(k)})$ and $(p_T^{(k)},p_S^{(k)})$ (based on their definition in Section 2.3.2).

Model	Parameter	Value
MLP	· # neurons in dense layers 1,2 and 3	2000
CNN	· Conv. kernel size for blocks 1, 2 and 3	(11,1), (10,1), (6,1)
	· Conv. siding stride for blocks 1, 2 and 3	(1,1), (1,1), (1,1)
	· # conv. kernels in blocks 1, 2 and 3	50, 40, 30
	· Pool. sizes for blocks 1, 2 and 3	(2,1), (3,1), (1,1)
	· # neurons in dense layer	1000
LSTM	· # LSTM cells in layers 1 and 2	64, 64
	· Output dim. of LSTM cells in layers 1 and 2	600, 600
	· # neurons in dense layer	512
Hybrid	· Conv. kernel size	(11,1)
	· Conv. sliding stride	(1,1)
	· # conv. kernels	50
	· Pool. size	(2,1)
	· # LSTM cells in layers 1 and 2	27, 27
	· Output dim. of LSTM cells in layers 1 and 2	600, 600
	· # neurons in dense layer	512
AE	· # neurons in encoder & decoder dense layer	5000