How can I see the connection of the hidden layer in CNN?

Hi,

in a CNN like in the example "Create Simple Deep Learning Network for Classification"

there are hidden layer built by the function convolution2dLayer. If in the first convolution layer the filter depth is for example 8, then we have 8 feature maps in the first layer. So I wounder how the perceptrons in the second layer are connected to the the perceptron of the first layer if the second layer has equal number of features (here in the example 8), more or less features? Could I adjust the connection within the parameter of the function convolution2dLayer?

Kind regards,

Clemens

Answer 1

[Mahesh Taparia]

Hi

I think you are asking how the CNN works. Let the size of the feature of the 1st layer is M1XN1XD1. For the second layer, let the size of kernel is pXqXD1 and there are D2 number of kernels, then size of features of 2nd layer will be some M2XN2XD2. So basically for 2D convolution, the depth of the kernel is same as feature depth and the it is sliding convolutional filter to the input. You can adjust the kernel size, number of kernel, padding, etc in convolution 2D layer. For more information, you can check the documentation of Conv 2D layer.