We now have the perfect solution for celebrity obsession: an algorithm that conjures up new famous faces on demand.
Researchers at Nvidia created the celeb-generating algorithm using a clever new machine-learning technique. The faces are dreamed up using a more efficient version of what’s known as a generative adversarial network (or GAN).
A GAN consists of two neural networks, both trained using a particular data set. One network then tries to generate synthetic examples to fool the other network into thinking they came from the original data set. The process helps the first network improve its ability to produce realistic data.
GANs were invented by Google researcher Ian Goodfellow (who is also one of our 35 Innovators Under 35 for 2017), and they have proved remarkably effective for synthesizing realistic-sounding speech and all sorts of dazzling imagery. They could prove very useful for generating animated graphics for video games, and for compressing video more efficiently.
In a paper (PDF) submitted to an upcoming conference, the Nvidia researchers claim to have developed a better GAN by having it start off working with low-resolution images, and gradually increasing the image resolution as well as the size of the networks involved. They fed their GAN a data set of celebrity faces, and it produced some very realistic-looking faces (you can check out a video of the research here).
One thing to note, however, is that a few of the images feature strange artifacts and features, like a missing eyebrow or teeth in the wrong place—not exactly things that would get you a gig on reality TV. This goes to show that even if machine learning can produce amazing visual trickery, it lacks the deeper intelligence required to make sense of the real world.
Why Meta’s latest large language model survived only three days online
Galactica was supposed to help scientists. Instead, it mindlessly spat out biased and incorrect nonsense.
DeepMind’s game-playing AI has beaten a 50-year-old record in computer science
The new version of AlphaZero discovered a faster way to do matrix multiplication, a core problem in computing that affects thousands of everyday computer tasks.
A bot that watched 70,000 hours of Minecraft could unlock AI’s next big thing
Online videos are a vast and untapped source of training data—and OpenAI says it has a new way to use it.
Google’s new AI can hear a snippet of song—and then keep on playing
The technique, called AudioLM, generates naturalistic sounds without the need for human annotation.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.