卷积核如何在 CNN 中进行训练？
How does a convolution kernel get trained in a CNN?
在 CNN 中，卷积运算 'convolves' 输入矩阵上的核矩阵。现在，我知道了全连接层如何利用梯度下降和反向传播来进行训练。但是内核矩阵如何随时间变化？
如前所述here, in the Keras documentation，内核矩阵有多种初始化方式。但是，我有兴趣知道它是如何训练的？如果它也使用反向传播，那么有没有论文详细描述训练过程？
This post也提出了类似的问题，但没有得到解答。
Here you have a well explained post about backpropagation for Convolutional layer. In short, it is also gradient descent just like with FC layer. In fact, you can effectively turn a Convolutional layer into a Fuly Connected layer as explained here.
在 CNN 中，卷积运算 'convolves' 输入矩阵上的核矩阵。现在，我知道了全连接层如何利用梯度下降和反向传播来进行训练。但是内核矩阵如何随时间变化？
如前所述here, in the Keras documentation，内核矩阵有多种初始化方式。但是，我有兴趣知道它是如何训练的？如果它也使用反向传播，那么有没有论文详细描述训练过程？
This post也提出了类似的问题，但没有得到解答。
Here you have a well explained post about backpropagation for Convolutional layer. In short, it is also gradient descent just like with FC layer. In fact, you can effectively turn a Convolutional layer into a Fuly Connected layer as explained here.