“Resnet"
Deep Residual Learning for Image Recognition
人们通过实验确信,随着网络的加深,可以提高网络的学习能力,但是现实结果是当达到一定深度后出现了学习退化的现象。Resnet 考虑的是在一种极端情况下,深层次网络什么也没有学到,但至少表现能力不会比浅层网络要差,即恒等映射(Identity mapping)。
In this case ,the shortcut connections can simply perform identity mapping, and their outpus are added to the outputs of the stacked layers