Table 1.

Hyperparameter settings considered for DeepSynergy

Hyperparameter	Values considered
Preprocessing	norm; norm+tanh; norm+tanh+norm
Hidden units	[8192, 8192]; [4096, 4096]; [2048, 2048];
	[8192, 4096]; [4096, 2048]; [4096, 4096, 4096];
	[2048, 2048, 2048]; [4096, 2048, 1024];
	[8192, 4096, 2048]
Learning rates	10−2; 10−3; 10−4; 10−5
Dropout	no dropout; input: 0.2, hidden: 0.5

Note: All possible combinations of the presented hyperparameters were optimized via grid-search.