Table 3.

Parameters listed in the text, found by the RayTune library for each dataset (Broadband NIRS and HELICoiD), model (linear absorption and nonlinear scattering), and neural network training strategies (a) and (b), respectively.

Dataset and model	Neural network	Hidden layers	Network width	Learning rate	Activation function	Batch size
Broadband NIRS	Strategy (a)	2	35	$5.28832\times {10}^{-3}$	Leaky ReLU	128
Linear	Strategy (b)	2	16	$1.7495724\times {10}^{-3}$	Leaky ReLU	32
Broadband NIRS	Strategy (a)	1	62	$7.86158\times {10}^{-4}$	Leaky ReLU	32
Nonlinear	Strategy (b)	2	53	$5.94006\times {10}^{-4}$	ELU	32
HELICoiD	Strategy (a)	1	31	$6.64275\times {10}^{-3}$	ELU	512
Nonlinear	Strategy (b)	2	46	$3.52619\times {10}^{-4}$	ELU	32