Research Excellence

Knowing Faces

How the Fruit Fly Illustrates the Brain’s Capabilities

Face recognition technology is now used in many applications. Financial institutions use it as an electronic personal identifier, and governments apply surveillance technology to improve public safety. Although face recognition technology has been developed to an advanced level, it is still unclear how humans are able to recognize such a large amount of faces in a fraction of a second. For the past three years, Professor Charles F. STEVENS from the Salk Institute for Biological Studies has sought to answer this question.

How Fruit Flies Recognize Odors

Many people can recognize the model of a car solely by listening to the engine. The subtle differences in sound, including the frequencies and certain other functions, allow accurate identification of each engine. This is a skill that humans can develop with practice. “If cars can be recognized in this way, it may well be possible to create an algorithm to describe the process by which our brains compute external data to achieve the feat of face recognition,” Stevens suggests.

“The human brain may be too complex a processor for its behaviors to be observed directly.” Stevens first sought others’ insights into how fruit flies differentiate one odor from another and then used available data to describe the process.

\\  Many contemporary research techniques were totally unimaginable when I first started research. My latest work is made possible only by the discovery of many new findings.  \\

Stevens explains the basics: “When it comes across an odor, a fly encodes this odor using a grid of 50 different receptors. These receptors send information to another part of the brain called the mushroom body. In this process, 2,000 neurons are involved, and each takes a random sample of the firing rates from the 50 receptors. As a result of the competition among these 2,000 neurons, 50 will be chosen to fire signals to the brain. These 50 signals are fired in a very compressed way that allows the fly’s brain to recover all of the encoded information associated with this odor. If the fly comes across another odor, a different set of 50 will fire.”

The fly’s brain can also use what is called a compressed sensing technique. “Think about how our cell phone camera takes a picture. The camera tries to reduce the number of pixels required to save the picture. For example, if there is a plain white wall in the background of a family photo, the camera does not need to remember every pixel of the plain background because they are largely the same. When saving the picture, the phone does not simply remember each pixel on the wall, but also recalls that the wall is of a uniform color. This is similar to how a fly remembers an odor. The 50 signals fired from the neurons are given a tag that represents its context to reduce the amount of information that must be transmitted.”

How Human Brains Recognize Faces

The visual system works by taking whatever patterns it sees and encoding the patterns into the firing of each of the cells that receives this input. The primary visual cortex cannot recognize an object—rather, its task is to encode all of the information needed to replicate the object. Recognition is performed by face patches, which use a combinatorial code in which a large group of cells respond to exactly the same stimulus at different rates. Building on the work of his colleagues and his own study of the activities of face-recognizing neurons in humans, Stevens showed that for every face, the cells that fire spikes always fire at the same average rate and that there is an exponential distribution of firing rates.

\\  There is no way I can predict how we will be doing research in ten years’ time or what questions we will be asking. The technology is developing much faster than we could ever imagine.  \\

Referred to as the “maximum entropy distribution” by statisticians and information scientists, the way that cells respond to different faces by firing spikes at different rates allows the largest number of faces to be encoded with the smallest number of errors. “Through this finding, we can say that the human brain recognizes faces in the same way that the fly recognizes odors,” Stevens says.

“Three parts of the human brain—the olfactory system (where information is taken in), hippocampus (where learning takes place), and cerebellum (where motions are learned)—share similar features with the operation of the fruit fly’s brain when it recognizes odors. The hippocampus takes information from a small number of cells and expands that information over a larger number of cells. It then uses inhibition to reduce the number of cells that fire spikes, and in that way generates a tag. The cerebellum is a structure that captures information on motion and expands it into a gigantic number of cells, which together make up half of all the cells of the brain. Despite this huge expansion, only a few of the cells fire spikes. Those few are tagged and used for learning motion.”

Depiction of the cerebellum

The Human Brain at Work

Apart from the quest to understand face recognition in human brains, Stevens has spent recent decades trying to discern the rules that regulate the operation of the human brain. Many of his studies involved the observation of a variety of species, ranging from goldfish to fruit flies, to discover whether similar rules are followed in human brains.

“My colleagues look at the way that a fly does things and compare it with the way that Google updates its database. We find that the fly is at least as good as Google and in many ways it’s even better. If this is also how the human brain works, we may well wonder how evolution has designed us in such a way.”

Use of fruit flies

Fruit flies are a convenient tool in research because their genetic information is relatively simple compared with many other animals. Every piece of their genetic information is known and can be manipulated. Edited genes can be inserted into fruit flies, after which their behavior can be observed. Flies for experimental use can be tagged at a stable position while the fly’s external environment is simulated, in a manner similar to virtual reality.

This research technique was only developed in the past decade and has greatly enhanced the capacity of Stevens’ research. Although he does not currently work with flies himself, the findings of his fellow researchers have already shed much light on the topic.

How Do Different Species Smell?

Each species has a different amount of odor receptors and their olfactory capability varies.

About Prof. Charles F. STEVENS

Prof. Charles F. STEVENS is currently a professor at the Molecular Neurobiology Laboratory at the Salk Institute for Biological Studies. His long-term research goal is to understand the mathematical operations carried out by neural circuits, and to understand the design principles that underlie the scalable architecture of neural circuits.

Most humans can rapidly learn to recognize any one of the seven billion human faces in the world, despite the fact that faces are extremely complex stimuli. Once a face has been learned, we remember it for a long time and can recognize a familiar face in a fraction of a second. In April, Stevens gave a lecture on “How the Primate Brain Recognizes Human Faces” at the IAS. He described where the neurons that encode faces reside in the brain, and explained the neuronal code the brain uses for face recognition.

Prof. Charles F. STEVENS
Molecular Neurobiology Laboratory, Salk Institute for Biological Studies