Aesthetic Computing

Aesthetic Computing (computational aesthetics) is a research field that is rapidly advancing due to the remarkable development of machine learning in recent years.

In the case of paintings, a large number of paintings are shown to people and evaluated for their level of "beauty," and the machine is given a combination of the painting and the evaluation as a "teacher" and then trains an unknown painting to see how well it matches the human evaluation of beauty. Most of the time, supervised learning is used. The picture is the explanatory variable and "beauty" is the response variable. We try to match these variables and make the system as Gap-free as possible. If too much, it may over-learning.

However, it is a little difficult to determine what kind of pictures to give.
When I see a paper on unsupervised learning that says, "We gave them a lot of beautiful pictures to learn", I wonder how they judged that picture was "beautiful"..... I am curious about this point because I have been told many times (-:

On the other hand, there is also an approach to investigating "beauty" from the viewpoint of factors such as coloring, composition, etc.
It is called Affective Engineering. It is a relatively new field of research that originated at MIT in 1995.

What made me feel relieved when I heard this kind of talk about machines and "beauty" was that "music composed by AI that has been trained to learn beautiful music is not beautiful at all." 

Beauty is quite a tough subject.。