When artificial intelligence is tasked with visual recognition of objects and faces, it assigns specific components of its network to face recognition – just like the human brain.
It seems that the human brain cares a lot about faces. It is dedicated to a specific area of recognizing them, and the neurons there are so good at their job that most of us can easily recognize thousands of individuals. With artificial intelligence, computers can now recognize faces with equal efficiency – and neuroscientists at MIT’s McGovern Institute for Brain Research have found that a computational network trained to recognize faces and other objects can solve them all. Surprisingly detects brain-like tactics.
Visualization of a preferred incentive eg face-rank filter. Whereas the filters in the early layers (eg, Transform 5) were maximally activated by simple features, the filters reacted to features that were part of the face (eg, the nose and eyes) in the mid-level layers. are visible (for example, Transform 9) and appear to represent faces in a more holistic way in late convolutional layers. Credits: Courtesy of Kanvisher Lab.
search, reported in science advancesuggests that millions of years of evolution with shaped circuits in the human brain have adapted our systems for facial recognition.
“The solution to the human brain is to separate the processing of faces from the processing of objects,” explains Katharina Dobbs, who led the Walter A. The Rosenblyth Professor led the study as a postdoc in the lab of McGovern investigator Nancy Kanwisher. , The artificial network he trained did the same. “And this is the same solution we envisage for any system trained to recognize faces and classify objects,” she adds.
“These two completely different systems have figured out what – if not – the good solution is. And it seems very deep,” says Knavisher.
functionally specific brain region
More than 20 years ago, Kanwisher and his colleagues discovered a tiny spot in the temporal lobe of the brain that responds specifically to faces. This area, which they named the fusiform face area, is one of several brain regions Kanwisher and others found devoted to specific functions, such as detecting written words, perception of vocal songs, and understanding language.
Kanwisher says that as she discovered how the human brain is organized, she has always been curious about the reasons for that organization. Does the brain require special machinery for facial recognition and other functions? “Why ‘Questions’ Are Hard in Science,” she says. But with a sophisticated type of machine learning called deep neural networks, his team can at least figure out how a different system would handle a similar task.
Dobbs, leader of a research group at Justus Liebig University Giesen in Germany, assembled hundreds of thousands of images to train a deep neural network in face and object recognition. The collection included the faces of more than 1,700 different people and hundreds of different types of objects, from chairs to cheeseburgers. All these were presented on the network, about which there was no clue. “We never told the system that some of them are faces, and some are objects. So it’s just a huge task,” Dobbs says. “It needs to recognize a face and recognize a bike or a pen.”
As the program learned to recognize objects and faces, it organized itself into an information-processing network consisting of units dedicated exclusively to facial recognition. Like the brain, this specialization occurred during the later stages of image processing. In both the brain and artificial networks, the early stages of facial recognition involve the more general vision processing machinery, and the final stages rely on face-dedicated components.
It is unknown how the face-processing machinery is generated in the developing brain. Still, based on their findings, Kanwisher and Dobbs say that the network does not require an intuitive face-processing mechanism to gain that expertise. “We haven’t made anything face-ish in our network,” says Kanvishar. “The networks managed to differentiate themselves without any face-specific nudges.”
Kanwisher says it was thrilling to see deep neural networks being separated into different parts for face and object recognition. “That’s what we’ve been seeing in the brain for 20-some years,” she says. “Why do we have a different system for facial recognition in the brain? This tells me that it is because it looks like an optimized solution.”
Now, he is eager to use deep neural nets to ask similar questions about why other brain functions are organized the way they do. “We have a new way of asking why the brain is organized the way it is,” she says. “Will the structure we see in the human brain be generated automatically by training networks to perform comparison tasks?”
Written by Jennifer Michalovsky
Source: Massachusetts Institute of Technology
