Artifical Intelligence / Machine Learning: Generative Adversarial Networks (GANs) / Generative Modeling

While there is a lot of buzz on data analysis and understanding data there for applied use of artifical intelligence and machine learning it is less known in the public on what Artifical Intelligence can do regarding generating original content and altering existing content.

So called Generative Adversarial Networks can generate or change pictures and can even alter videos. E.g.

The website www.ThisPersonDoesNotExist.com gives and interesting and somewhat worrisome examples of that technolgy.

The speed at which this technology evolves is as astounding as the results it can give already.

The technology can not only be used to generate static content like the pictures of faces, but can also modify videos. There is an example where the same scene filmed by day is recreated to appear at night:

Another example illustrates how a video of a horse is transferred to a video of a zebra. As Ian Goodfellow points out the GAN did not only change the brown horse to be white with black stripes, but also adjusted the background to appear more brown than green. Likely because the data of zebras contained more African-like climate zones as opposed to the lush green surroundings in the original video.


Sources:

Comments

Popular posts from this blog

SQL Server Setup: Windows Firewall warning (Ports)

SQL Server 2014 Enterprise Edition: Server Setup: Feature Selection

How to read an assembly.dll.config