In October 2018, a division of Google called DeepMind created a neural network BigGAN for inventing photos. NVIDIA went ahead and used a similar algorithm to create faces that don’t exist. The result of the neural network shown in the video — it is both impressive and scary. The fact that computer-generated portraits are no different from the present, and there is hardly a person who can immediately recognize a fake.
As with many such projects, the developers used the generative-adversarial neural network GAN. One part acts as a generator of images, and the second evaluates their similarity with samples of the ideal result. As samples were used celebrity photos from the set CelebA-HQ. It consists of approximately 30 000 images, so the lack of a training base of the developers clearly was not.
To achieve “unprecedented quality,” the neural network is first created images in low resolution. With the improvement of the results of its work, the programmers have added new layers with higher resolution. The process of creating the images you can see in the video below, and to read technical document in PDF format.
The development of artificial intelligence frankly scary. Neural network NVIDIA is not the only system that reflects something human. In may 2017 was presented to the project WaveNet from DeepMind, perfectly mimicking human speech. Also, the developers are actively developing robots capable of understanding and expressing human emotions. A striking example — the robot of Sofia, was developed by company Hanston Robotics and obtained citizenship of Saudi Arabia.
What do you think, what else can an artificial intelligence? Your bold assumption, write in the comments or in our Telegram chat.