Programmer showed how neural network forgets the face of man

At first it “grows old”, and then turns into something terrible, until it completely disappears.

The programmer first generated a “perfect” human face, and then made the neural network forget it. The algorithm turned off the neurons one by one, until even the main features disappeared. The developer’s video was noticed in Vice.

To generate face, the programmer used adversarial neural networks that compete with each other to produce a realistic result. The algorithm was trained on millions of portraits, so its individual neurons took into account the main features of a person’s face: eyes, skin color, shape, hair, and not only.

After that, the author of the project programmed the system to disconnect the neuron behind the neuron until the entire network “forgets” what a person looks like. From the side, the result turned out to be creepy: first, the face began to “grow old”, then it turned green, and then only the basic features remained. After a minute, the neural network forgot the image completely.

In a conversation with Vice, the author of the project said that she was inspired by “human perception.” According to her, all that people see is a brain interpretation of the surrounding world, and people do not have access to “external” reality.

The brain creates a different reality, but no one can call it a bluff. Neuroscientists tell us that deep neural networks are similar to the visual system in some aspects. Therefore, for me, this is a unique opportunity to see the change in the world in someone’s consciousness, albeit artificial.the author of the experiment

She compared her video with the work of Claude Monet, who as he grew older became more blurred and acquired green and yellow shades. The girl noted that the human brain and eyes over time undergo changes that people are barely able to notice.

Back to top button