Neuroscientists from the McGovern Institute for Brain Research at MIT and Germany's Giessen University
Over 20 years ago Nancy Kanwisher, researcherfrom MIT, and her colleagues found a small area in the temporal lobe of the brain that specifically responds to faces. Scientists have named this area the spindle-shaped area of the face. Neuroscientists have shown that this part of the gyrus is responsible for face recognition. However, until now, scientists do not know what is the reason for such a specific allocation of individual objects.
As Kanvischer notes, in the new study, they wanted to test how another system wouldNeuroscientists collected hundreds of thousands of images to train a neural network.The collection included images of the faces of 1,700 different people and hundreds of objects (fromThe whole set was presented by a neural network without any prompts.
“We didn’t tell the system that some of theimages are faces, and some are other objects. We had one big challenge,” says Katharina Dobs, co-author of the study from the University of Giessen. “AI should recognize a face in the same way as a bicycle or a pen.”
The researchers saw that when the programlearned to identify objects and faces, she organized herself into an information processing network. The network has formed separate blocks specifically designed for face recognition. As in the human brain, this specialization occurs in the late stages of image processing: first, the general mechanisms of vision are used, and at the last stage, the components responsible for face recognition are connected.
Neuroscientists note that networks trainedonly objects perform poorly at face recognition and vice versa, and networks optimized for both tasks spontaneously split into separate systems for faces and objects. Such a division, according to scientists, is fully consistent with their observations of the work of the human brain.
"The human brain decides to separate the processing of faces from the analysis of other objects," saysDobs.—The artificial grid has done the same thing.Any system that has been trained to recognize faces and other objects."
Researchers believe that if both nature andthe neural network came to the same principle of operation, such a solution is optimal. They plan to use machine learning to figure out why other brain functions work the way they do.
Read more:
There is another “planet” inside the Earth: how it saved nascent life
New study refutes light energy transfer theory
Everything you knew about the tyrannosaurus is not true: how science is changing its Hollywood image