In Introduction to tensorflow course, Week3,Video2(What are convolutions and pooling), How do you decide the value of the filters? Is it based on the final objective that we need to get? Or is it based on a set of rules or procedures? Can somebody clear this doubt?
You initialize them randomly for “symmetry breaking” and then useful values are learned through training of the network on your training data. If you are using TF, that initialization happens automatically, but there are some choices for which initialization algorithm to use that you can configure. I’m not familiar with the order in which they present the material in the TF specialiations, so “hold that thought” and it will be explained as you go through the courses.
If that is news to you, it might be a good idea to consider taking the DLS specialization before you take the TF specialization. I think that in the TF courses they pretty much assume you already know what a fully connected network or a convolutional network are and just explain how to implement them in TF. DLS teaches the fundamentals of how the different types of networks work and what types of problems they are useful for solving and introduces you to TF.