具体推导可参见网页:
http://math.stackexchange.com/questions/945871/derivative-of-softmax-loss-function
导数推导基本公式:
http://www.docin.com/p-424329540.html
http://wenku.baidu.com/link?url=AGwmcbk0zpIOjT7a1Mx7dIGLZlYDRUkug6_6BwbOsgKEdxJ4HLuPMqvMoXWGuDjhjAzz0MdUQM3YIK2nSFgPnl3XirBqCp4coKcd1yFDHDa