深度学习优化- Deep Residual Learning 残差结构

1. Purpose

Achieve a balance between depth layer of neural networks and complexity, as well as gain accuracy.

2. Stucture
   Basic Residual Block ：
   Instead of hoping each few stacked layers directly fit a desired underlying mapping.
   Shortcut connections are those skipping one or more layers. In this case, shortcut connection perform identity mapping, and their outputs are added to stacked layers.

highlight:Identity shortcut connections add neither extra parameter nor computional complexity.