Training cores of convolutions of a neural network?

Question

Neural networks

I want to make a convolutional neural network, I watched various videos / articles and the following question: (1) the values ​​​​of the convolution kernels are selected randomly and set to optimal values ​​​​during the program, or (2) these values ​​\u200b\u200bmust be written manually (already prepared kernels with affixed values) ?
Somewhere they said that the (1) option is used and the neuron itself calculates the necessary values, but no matter how much I searched, I can’t find information about training the convolution kernel, and besides, they use static kernels everywhere that do not change their values.
If there are any articles on this case - do not hesitate to throw it off: D

I should have 7 bundles of 5x5 in size in the National Assembly and somehow I don’t really want to write them manually)) I am doing
the National Assembly according to the Wikipedia articlehttps://ru.wikipedia.org/wiki/%D0%A1%D0%B2%D1%91%D...
And here is the architecture itself:
https://upload.wikimedia.org/wikipedia/commons/5/ 5...

Answer 1

You only set the network configuration: size and quantity. And even these parameters can be selected by enumeration.
An example of a finished grid
https://cs.stanford.edu/people/karpathy/convnetjs/