How to generate images using a neural network?

Question

Neural networks

I figured out how to use neural networks to select and classify objects, but I don’t understand at all how to generate images with their help.
Here is an example. Suppose there is a simple neural network that, after training, is able to distinguish a circle from a triangle. If we give it an image of a circle as input, we will get zero at the output, if a triangle - one, and everything else (for example, a square or a triangle with rounded corners) is mapped into the range [0; 1] depending on what it is closer to - a triangle or circle.
And now how to make it work in the opposite direction, so that after receiving a number from 0 to 1, the neural network returns some image close to a triangle or a circle, depending on the number?

Answer 1

You need to train not only the recognizer, but also the generator - which takes your 1 and generates a triangle out of it. Somewhere I saw an explanation of all this, maybe even on Habré.