全连接层、卷积层和池化层的正向及反向计算参考如下两篇文章:
https://www.zybuluo.com/hanbingtao/note/476663
https://www.zybuluo.com/hanbingtao/note/485480
softmax和交叉熵相关推导参考如下文章:
https://blog.csdn.net/weixin_49374896/article/details/123153552
Lenet5各层的详细参数配置参考:
1. 论文《Gradient-Based Learning Applied to Document Recognition》
2.https://cloud.tencent.com/developer/article/1931721
VC程序示例下载:
https://download.csdn.net/download/ggw007/85837600
其他一些推导过程:
整体架构:
损失函数:
激活函数和误差反向传播:
池化层: