How much w and b values are adjusted in every trial of functioning? Is there any effective number of networks?

Question

I am having trouble with understanding deep learning. I see that deep learning is basically about inductive process, and so the function must be adjusted enough until it hit