NNFM with varying training size

We varied the training data percentage from 5%-95% to test the performance of the neural network fitting the forward problem of diffuse reflectance spectroscopy. With each training data size, the neural network was trained independently for 20 times. Total normalized mean absolute errors over all source detector distances are recorded.

我们将训练数据占比设置为5%至95%不等，以测试神经网络拟合漫反射光谱（diffuse reflectance spectroscopy）正向问题的性能。针对每一种训练数据规模，我们均独立开展20次神经网络训练，并记录所有源探测器距离下的总归一化平均绝对误差。