Dropout

Posted on Edited on In

Dropout Layer

To avoid two main defects in training deep network, overfit and time consuming.

It has the ability to achieve regulation.

The process of Dropout layer is that hide a part of the node i one layer and start forward and backward. So only a part of nodes that didn’t be hidden are updated. Repeat this process.

Reference

https://zhuanlan.zhihu.com/p/38200980