In Week 4, an intuition about deep representation was discussed. It was explained that every layer on the deep network represents a smaller problem scope. As mentioned in the facial detection example, layer 1 was used to detect the “edges” of an image. Layer 2 is to detect “eyes”, “mouth”, etc. and Layer 3 is to detect “facial patterns”. How does layer 1 detect the edges on the images concretely? If we add more hidden layers, which smaller scope is it intended to detect? How do we explicitly indicate that his layer is used to detect this smaller subset of the problem?
Hi, Svillasica!
Welcome to the community.
The deep learning intuition that you are talking about is based on layer networking. In Course 1, Prof. Ng talks about neural networking in the layers i.e how we are building the network from one layer to another with the right input features. The assignment itself represents the whole illustration and explanation if you visit that again.
This was a general representation of what is going on. In the next illustration, it talks about the same in-depth.
Well, I know that you are very curious to know about the intuitions, the model is trained at. For that, you have to wait a bit. Prof. Ng will make you clear in his later courses on how you detect the edges to get the perfect output. I am giving you a general idea on what things you will learn later.
You will use filters, strides and padding like built-in platforms to detect the edges of an image at a proper place. Hope, this will ease out your eager
