Human Facial Recognition: Move the Method to Software, Right?
January 23, 2020
Here is a unique use of AI. Discover reports, “Algorithm Accurately Reconstructs Faces from a Monkey’s Brain Waves.” The write-up cites a study from the Caltech that took researcher Doris Tsao and her team nearly 15 years to complete. The research was performed on macaque monkeys because their ability to differentiate faces is similar to that of humans. Writer Nathaniel Scharping tells us:
“Working with macaques, the researchers say they identified a small group of neurons that are specialized to pick out individual features of faces and assemble them into a single image. With an fMRI machine, they watched as the monkeys looked at a series of 2,000 different faces and recorded which neurons were active in the so-called ‘face space’ of their brains. They could even pinpoint specific neurons that corresponded to different features. In all, they assembled a library of about 200 neurons that work to piece together facial features. The real test of their work, however, came when they worked backwards, reassembling faces using only information from firing neurons. With data on how the facial feature neurons fired, they rebuilt uncannily accurate versions of the faces. You can see the resemblance in the image above, from the study published Thursday in Cell. The reconstructed faces look almost exactly like the faces seen. The image is stark proof that primate brains’ break faces apart into tiny, digestible pieces as part of the process of recognition.”
The article shares the original and reconstructed images, and they are indeed nearly identical. The researchers have a lot more work ahead of them; they had previously pegged six areas of the brain responsible for processing faces, and this experiment only looked at two of them. Tsao and her team hope their work might lead to help for the blind. By stimulating the right neurons, someday we may be able to communicate visual information right into the brain, bypassing the eyes altogether.
Cynthia Murrell, January 23, 2020