A Novel Deep Learning Approach for Recognizing Stereotypical Motor Movements within and across Subjects on the Autism Spectrum Disorder

Table 2

Experimental setup for the CNN model.

Parameter	Value	Parameter	Value
Size of input vector	50 (frequency) - 90 (time)	Learning rate	0.01
Number of input channels	9	Weight decay	0.0005
Number of feature maps	24~500	Momentum	0.9
Filter size	7x1 ~ 11x1	Dropout	0.5
Pooling size	3x1	Size of mini-batches	150
Activation function	ReLU	Number of epochs	10~40