Hello, why is the pooling layer placed before the convolution layer in downsampling, while I've learned that it's usually placed after? What advantages does this approach offer?

[magic524]