Deep learning consists of ConNN layers followed by pooling layer and MLP NN Such networks mainly are mainly applied to find the features of $3$D input images at different levels; low level, medium level and high level.

For $3$D input image construct such network that deals for the three levels. At each network layer assign the kernels that guarantee mapping $3$D input to $3$D output.

For m input maps of layer $(b-1)$ assign the cubic filters that are used to get $D$ maps at layer $(b_{{?}_{?}}).$

How to get any layer $(b_{?})$ for m maps of layer $(b-1).$ Demonstrate by analysis that supported by mathematics.

To learn layer $(b-1)$ you need to update the weights of layer $(b),$ how to achieve that considering that any map at layer $(b)$ is due to the sum effect of all maps at layer $(b-1).$