The Surprisingly Simple Way Your Brain Recognizes Faces
Scientists can reconstruct entire faces with neurons.
When you see a familiar face, your brain recognizes who it belongs to in an instant. And while you might assume that it would be a complicated cognitive process to scan someone’s facial features and retrieve every little smile line and eyebrow hair, it’s actually quite simple.
In a study published Thursday in the journal Cell, Caltech neuroscience researcher Doris Tsao and post-doctoral researcher Steven Le Chang show how a tiny group of neurons recognize faces, unlocking the code to how brains quickly identify faces. Here’s how it works.
There is a small part of a monkey’s brain that specifically identifies faces. By examining the action of 205 neurons in this region, researchers found that they could do two things: They could predict which neurons would fire when a monkey viewed a particular face, and even more remarkably, they could digitally reconstruct a face the monkey was viewing just by looking at which neurons were firing.
Chang and Tsao could predict what face a monkey was looking at by measuring action potentials at a group of neurons called "face cells."
Tsao had previously collaborated on a project to identify a region of the rhesus macaque’s brain, in the temporal lobe, that was exclusively used for identifying faces. In the Cell, study, she again focused on this area, specifically the middle lateral (ML)/middle fundus (MF) and anterior medial (AM). Neurons in these areas fired especially strongly when processing faces; scientists call them “face cells.” Tsao says this discovery has made it possible to design experiments to explore facial identification.
“Since we discovered face-selective regions in monkeys almost 15 years ago, we’ve been trying to understand the precise code used by cells in these regions to encode facial identity,” Tsao tells Inverse.
Researchers looked at three previously identified areas of rhesus macaques' brains to figure out what the brain is doing when it identifies faces.
The researchers hypothesized that, rather than encoding for each individual face that a monkey sees or recognizes, it’s more likely that these cells in the brain divide the face into specific regions and measure each of these dimensions. These include things like distances between eyes, heights of eyebrows arches, and different hairline configurations — traits that our brains subconsciously process.
The monkeys, which were trained to focus on a particular part of a screen in exchange for juice rewards, sat in a fMRI machine while images of faces and other non-face objects flashed in front of their eyes. Electrodes read the strength of action potential firing in the facial recognition region of the monkeys’ brains, and the scientists recorded these signals. The researchers associated these signals with 50 different facial dimensions, and in doing so, were able to reconstruct the faces the monkeys were seeing by using the monkeys’ neuron firing patterns as a map.
They could do this because the way the brain codes facial identity is so simple. Evolutionary biologists suspect that quick facial recognition was key for the survival in our ancestors. But once that identification happens, things get a bit fuzzy.
“Now that we have access to this high-level code, we can address many new questions,” Tsao says. “For example: How does the brain use this code to make various decisions about faces, downstream of face patches? Does the brain use the same code to remember and imagine faces, and if so, how?”
Abstract: Primates recognize complex objects such as faces with remarkable speed and reliability. Here, we reveal the brain’s code for facial identity. Experiments in macaques demonstrate an extraordinarily simple transformation between faces and responses of cells in face patches. By formatting faces as points in a high-dimensional linear space, we discovered that each face cell’s firing rate is proportional to the projection of an incoming face stimulus onto a single axis in this space, allowing a face cell ensemble to encode the location of any face in the space. Using this code, we could precisely decode faces from neural population responses and predict neural firing rates to faces. Furthermore, this code disavows the long-standing assumption that face cells encode specific facial identities, confirmed by engineering faces with drastically different appearance that elicited identical responses in single face cells. Our work suggests that other objects could be encoded by analogous metric coordinate systems.
