# Designing Cities and Furnitures With Machine Learning | Two Minute Papers #36

## Метаданные

- **Канал:** Two Minute Papers
- **YouTube:** https://www.youtube.com/watch?v=kMa_B3wLxAM
- **Дата:** 09.01.2016
- **Длительность:** 2:33
- **Просмотры:** 8,735

## Описание

Creating geometry for a computer game or a movie is a very long and arduous task. For instance, if we would like to populate a virtual city with buildings, it would cost a ton of time and money and of course, we would need quite a few artists. This piece of work solves this problem in a very elegant and convenient way: it learns the preference of the user, then creates and recommends a set of solutions that are expected to be desirable. The weapon of choice to accomplish this was Gaussian Process Regression.

___________________________________

The paper "Interactive Design of Probability Density Functions for Shape Grammars" is available here:
http://lgg.epfl.ch/publications/2015/proman/index.php

The thumbnail image was created by See-ming Lee (nice name, btw!) (CC BY 2.0) - https://flic.kr/p/oewqwn

Subscribe if you would like to see more of these! - http://www.youtube.com/subscription_center?add_user=keeroyz

Splash screen/thumbnail design: Felícia Fehér - http://felicia.hu

Károly Zsolnai-Fehér's links:
Patreon → https://www.patreon.com/TwoMinutePapers
Facebook → https://www.facebook.com/TwoMinutePapers/
Twitter → https://twitter.com/karoly_zsolnai
Web → https://cg.tuwien.ac.at/~zsolnai/

## Содержание

### [0:00](https://www.youtube.com/watch?v=kMa_B3wLxAM) Segment 1 (00:00 - 02:00)

Dear Fellow Scholars, this is Two Minute Papers with Károly Zsolnai-Fehér. Creating geometry for a computer game or a movie is a very long and arduous task. For instance, if we would like to populate a virtual city with buildings, it would cost a ton of time and money and of course, we would need quite a few artists. This piece of work solves this problem in a very elegant and convenient way: it learns the preference of the user, then creates and recommends a set of solutions that are expected to be desirable. In this example, we are looking for tables with either one leg or crossing legs. It should also be properly balanced, therefore if we see any of these criteria, we'll assign a high score to these models. These are the preferences that the algorithm should try to learn. The orange bars show the predicted score for new models created by the algorithm - a larger value means that the system expects the user to score these high, and the blue bars mean the uncertainty. Generally, we're looking for solutions with large orange and small blue bars, this means that the algorithm is confident that a given model is in line with our preferences. And we see exactly what were looking for - novel, balanced table designs with one leg or crossed legs. Interestingly, since we have these uncertainty values, one can also visualize counterexamples where the algorithm is not so sure, but would guess that we wouldn't like the model. It's super cool that it is aware how horrendous these designs looks. It may have a better eye than many of the contemporary art curators out there. There are also examples where the algorithm is very confident that we're going to hate a given example because of its legs or unbalancedness, and would never recommend such a model. So indirectly, it also learns how a balanced piece of furniture should look like, without ever learning the concept of gravity or doing any kind of architectural computation. The algorithm also works on buildings, and after learning our preferences, it can populate entire cities with geometry that is in line with our artistic vision. Thanks for watching and for your generous support, and I'll see you next time!

---
*Источник: https://ekstraktznaniy.ru/video/14893*