Actress Kristen Stewart has co-authored a scientific work on artificial intelligence

News from the category of events that happen, perhaps, is not more likely than the events of the level of discovery of the Higgs boson. The popular Hollywood actress Kristen Stewart surprised the whole community of researchers in the field of artificial intelligence, becoming a co-author of the published scientific work on computer training.

The dream of millions of fans of the Twilight franchise recently made a directorial debut and shot the short drama "Come Swim", when working on it, the machine learning technique, known as "style translation", was used. This is a method in which the aesthetics of a single image or video frame are applied to the aesthetics of another image or frame to create an impressionistic visual style.

With the support of the visual effects specialist Bhaoutik Yoshi and producer David Shapiro, Stewart co-wrote a scientific paper describing the process of creating the film and the technologies that were used in it. The article is published in the largest scientific online library arXiv, where publications do not pass scientific review.

Fans of the actress, as well as experts in the field of artificial intelligence research, were surprised (and happy) to learn that Stewart took an active part in this work.

The work itself has the title "Revitalization of Impressionist paintings in the film Come Swim using style transfer based on neural networks" and offers a detailed analysis of how you can use this method of machine learning to create a movie. In the article, the short film "Let's Go for a Float" is described as "a poetic impressionistic portrait of a grief-stricken person under water".

To transfer the visual style between frames, the team of the filmmakers used existing neural networks, with which they first transferred the desired stylistics to the test frame, and then step by step changed the image, adding "blocks of new color and textures" until they received the desired Effect. After the desired style transition setting, they applied this method in different parts of the picture, creating frames, like the ones that can be seen below. In fact, this method is similar to some very advanced version of the progressive scan.

Of course, the expected negative reaction to the published work followed: "What the hell did the Hollywood actress get into the machine learning environment, where she does not understand a damn thing?" However, more important here is the fact that the tools of machine learning, which were once interesting only to those people who practically lived with them, are becoming more and more popular. Open sources of structural AI like the same Tensor Flow and Keras allow anyone to try their hand at writing AI code and creating on their basis specific commercial processing methods like the same style transfer (by the way, the same social network Facebook is very actively using image filters created by on the basis of this method) and promote the promotion of such technologies in culture.

In the end, the revolution of AI technologies will depend not just on the huge data sets and powerful processors that are able to classify and classify these data for work. It is also important to create an open community of developers, as well as available tools for work. And Stewart's scientific article is an excellent example of how this can work and what we can achieve through this.

The short film "Let's go swimming" was shown at the Sundance-2017 film festival. Below you can see the teaser movie.

The article is based on materials https://hi-news.ru/entertainment/aktrisa-kristen-styuart-stala-soavtorom-nauchnoj-raboty-ob-iskusstvennom-intellekte.html.