Rethinking the Usage of Batch Normalization and Dropout in the Training of Deep Neural Networks 阅读笔记与实战 05-22