TABLE 3.

Hyper-parameters of NanoReviser.

Layer	Feature source	Hyper-parameters	Value
CNN	Signal input	Kernel size	1 × 3
		Number of filters	8
		Stride	1
		Activation function	ReLUa
Bi-LSTM	Read input	State size	16
		Activation function	tanhb
Bi-LSTM	Concatenated input	State size	64
		Activation function	tanhb
Dropout	Concatenated input	Dropout rate	0.2
Center loss	Concatenated input	Proportion	0.2
Adam optimizer		Initial learning rate	0.002
		Decay rate	0.05
		Beta_1	0.9
		Beta_2	0.999

^aReLU is one of the commonly used activation functions in CNN layer. ^btanh is the default activation function in LSTM layer.