Understanding the difficulty of training deep feedforward neural networks, Xavier Glorot, Yoshua Bengio, 2010Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics, Vol. 9 (PMLR) - 介绍了Xavier(又称Glorot)初始化方法,一种通过保持跨层信号方差来初始化神经网络权重的方法,特别适用于Sigmoid和Tanh激活函数。