How to teach a neural network to analyze logos and draw?

Question

Neural networks

How to use machine learning to teach a neural network to process logos / images and draw something similar?
What libraries to use or where you can read about the interaction of neural networks with graphic elements, and not just with mathematical ones.

Answer 1

Read about GAN - generative adversarial network.
In general terms, there are two neural networks. One generates pictures, the other evaluates how well the first one does. For example, a lesson about GAN on TensorFlow (in English).
You can try to train the network on a large collection of logos. For example with Brands of the World . And see what this couple will do in terms of creating new, inspired ones.

interaction of neural networks with graphic elements, and not only with mathematical ones

Neural networks work only with "mathematical elements", with numbers. To feed a picture to the network, it is translated into an array of numbers: each pixel in 1 (gray) or 3 (red, green, blue) values ​​​​from 0 to 255.
Perhaps a more interesting approach would be without neural networks. Have you heard of "generative art"? When they write programs that create something beautiful. It would be possible to combine the experience of living designers with a large portfolio, formalize their approach to creating logos, and randomly (?) choose one of the paths. Such a wide decision tree: start with a graphical primitive (circle | straight line | curve of the Nth order), then add points to it, build from points (?) and so on, far) Entrust the assessment of what has already been created to the subjective view of a person.

Answer 2