A New Type of Eye Movement Model Based on Recurrent Neural Networks for Simulating the Gaze Behavior of Human Reading

Table 1

Hyperparameter settings.

Layer	Hyper-parameter	Value	Parameter Range
Embedding	output dim	32	/
	input dim	1200	/
	input length	60	/
CNN	kernel size	3
	maxpool size	3
	stride size	1	/
L-LSTM	units	32	/
R-LSTM	units	32	/
Time Distributed	units	50	/
Global	activation	ReLU	[Sigmoid, Tanh, ReLU]
	cost	cross_entropy	/
	kernel initializer	random_normal	/
	optimizer	SGD	[SGD, AdaDelta, Adam, RMSProp]
	learning rate	2.5
	learning epoch	100
	batch size	16
	validation split	0.2	/