In what range to generate initial kernels for a convolutional neural network?

Question

Neural networks

The question is quite simple, but for some reason I could not find the answer in Google. For example, for a multilayer perceptron, it is recommended to generate weights and biases in the range [-0.5; 0.5] , and in what interval should we generate initial kernels for a convolutional neural network?

Answer 1

Any non-zero