常见的激活函数及其特点:https://zhuanlan.zhihu.com/p/85971385
激活函数及其导函数图像 https://blog.csdn.net/weixin_45620656/article/details/105984307
激活函数的饱和性:https://www.cnblogs.com/tangjicheng/p/9323389.html
梯度弥散与梯度爆炸:https://www.cnblogs.com/yangmang/p/7477802.html
偏置值反向求导公式推导:https://xinliu.blog.csdn.net/article/details/114503205
http://neuralnetworksanddeeplearning.com/chap5.html#the_vanishing_gradient_problem
BP算法,用梯度下降法更新权值W与偏置项b https://blog.csdn.net/caomin1hao/article/details/102323942