Bees Can Learn to Recognize Human Faces
CLAYTON, Victoria, Australia, February 5, 2010 (ENS) – Discoveries by bee researchers from Australia and France could lead to improved artificial intelligence systems and computer programs for facial recognition, the scientists said.
“What we have shown is that the bee brain, which contains less than one million neurons, is actually very good at learning to master complex tasks. Computer and imaging technology programmers who are working on solving complex visual recognition tasks using minimal hardware resources will find this research useful,” said Dr. Adrian Dyer of Monash University.
Dyer and his team of researchers individually trained different groups of free flying bees with a sugar reward for making correct choices, or alternatively the bees were punished with a bitter tasting solution for incorrect choices.
Bees can learn to recognize faces. (Photo by Umberto Salvagnin)
Faces were presented on a vertical screen and bees slowly learned to fly to the correct target faces.
Over the course of a day a bee brain learned a complex task, Dr. Dyer explained. Then, when tested in non-rewarded tests, only bees that had experience of multiple views of faces at both zero degrees and 60 degrees were able to solve a novel rotational angle of 30 degrees.
The study, performed over two years in Australia and Germany by Dr. Dyer with the support of the U.S. Air Force Office of Scientific Research, and Dr. Quoc Vuong from Newcastle University, UK, was published in the science journal “PLoS ONE” in January 2009.
But at the Universite de Toulouse, France Professor Martin Giurfa suspected that that the bees were not really learning to recognize people.
“Because the insects were rewarded with a drop of sugar when they chose human photographs, what they really saw were strange flowers. The important question was what strategy do they use to discriminate between faces,” explained Dr. Giurfa.
Wondering whether the insects might be learning the relative configuration of features on a face, Giurfa contacted Dyer and suggested that they go about systematically testing which features a bee learned to recognize to keep them returning to Dyer’s face photos.
The team published their discovery that bees can learn to recognize the arrangement of human facial features on January 29, 2010 in the Journal of Experimental Biology.
Bees hold no fear for this woman. (Photo by Monika)
Working with Universite de Toulouse PhD student Aurore Avargues-Weber, the team first tested whether the bees could learn to distinguish between simple face-like images.
Using faces that were made up of two dots for eyes, a short vertical dash for a nose and a longer horizontal line for a mouth, Avargues-Weber trained individual bees to distinguish between a face where the features were cramped together and another where the features were set apart by rewarding it with a weak sugar solution.
She then tested whether the bee recognized the pattern by taking away the sugar reward and waiting to see if the bee returned to the correct face. It did.
So the bees could learn to distinguish patterns that were organized like faces, but could they learn to categorize faces? the researchers asked. Could the insects be trained to classify patterns as face-like versus non-face like, and could they decide that an image that they had not seen before belonged to one class or the other?
To answer these questions, Avargues-Weber trained the bees by showing them five pairs of different images, where one image was always a face and the other a pattern of dots and dashes.
Bees were always rewarded with sugar when they visited the face while nothing was offered by the non-face pattern.
Having trained the bees that face-like images gave them a reward, she showed the bees a pair of images that they had not seen before to see if the bees could pick out the face-like picture. They did.
The bees were able to learn the face images, not because they know what a face is but because they had learned the relative arrangement and order of the features, the researchers concluded.
But how would the bees cope with more complex faces? the researchers asked.
This time, the team embedded the stick and dot faces in face-shaped photographs. Would the bees be able to learn the arrangements of the features against the backgrounds yet recognize the same stick and dot face when the face photo was removed? They learned.
When the team tried scrambling real faces by moving the relative positions of the eyes, nose and mouth, the bees no longer recognized the images as faces and treated them like unknown patterns.
So bees do appear to be able to recognize face-like patterns, but this does not mean that they can learn to recognize individual humans, the scientists concluded. They learn the relative arrangements of features that happen to make up a face-like pattern and they may use this strategy to learn about and recognize different objects in their environment.
“We discovered that bees distinguish face-like configurations. Bees learned through experience to look for a specific configuration of the different features, eyes, nose and mouth,” Dr. Dyer said.
“Although human faces aren’t important to bees, that they were able to recognize configurations and integrate visual features suggests that they do have an ability to reliably recognize very complex images, even in noisy or cluttered conditions.
“These are currently very challenging tasks for machine vision,” said Dr. Dyer, “and perhaps we can learn some tricks from the bees for teaching robots.”