Whether it’s old gym clothes, a wet dog, or strong body odor — our brains are remarkably good ignoring pervasive smells. It’s a quirk of our olfactory system that’s called habituation, which increases focus on new and threatening smells. Beyond uses in our brain, scientists believe a form of habituation can be used by A.I. to process massive amounts of data.
Borrowing neural circuitry from a fruit fly, scientists have designed an algorithm to mimic this neurobiological phenomenon, hoping to learn more about habituation. Researchers think this algorithm could design an electronic nose to sniff out pollution in cities or chemical weapons in combat.
Even if you’ve got the best nose in the world, it’s very likely that there are smells your nose has hidden from you.
Take your house for example. To you, your house might not smell like anything, but a person visiting for the first time might quickly pick-up on the fact that you have a not-so-small addiction to lavender lotion. This discrepancy isn’t due to anything fundamentally wrong with your sniffer, but rather that your brain has had enough prolonged exposure to the scent to decide it’s not important enough to bring to the forefront of your attention. Unlike, say, a sudden gas leak in your kitchen that would pose a danger.
In new research, published Monday in the journal Proceedings of the National Academy of Sciences, a group of researchers from UC San Diego, the Salk Institute for Biological Studies and Cold Spring Harbor Laboratory draws on the pastoral example of a dog in a field of flowers to illustrate this point.
“For example, if a dog sitting in a garden habituates to the smell of the flowers, then any change to the environment — for example, a coyote appearing in the distance — would more likely be detected by the dog, despite the fact that the coyote’s smell represents only a small component of the raw odor entering the dog’s nose,” write the authors.
Looking beyond mammals, the researchers chose to focus on an old, scientific standby, co-author and associate professor of biology at Cold Spring Harbor Laboratory, Saket Navlakha, tells Inverse: fruit flies.
“Fruit flies are a wonderful model because so much is known about their neural anatomy and physiology. Fruit flies have about 100,000 neurons in their brain and recently, a large portion of their connectome was mapped out at very fine (synaptic) resolution,” A connectome is a comprehensive map of neural connections in the brain, think of it like a wiring diagram for the most complex computer in the world.
“This has made it possible to uncover the algorithms that neural circuits have evolved to solve basic neurobiological problems. In addition, there is evidence that some of the basic computational strategies used by fruit flies may be conserved in mammals; so studying fruit flies may give us a basis for understanding computations in other species.”
After frolicking in a field of flowers for awhile, a dog will become habituated to the scent and stop noticing it. This is important because it allows them to notice other scents, like a dangerous coyote. PNAS
In their algorithm, the researchers mimicked a part of the fruit fly’s neural circuitry called the “negative image” model of habituation. Essentially, this model works by storing a habituated odor and comparing it with new odor landscapes to detect differences or new smells. In animals, like fruit flies, this helps differentiate between new smells that could be dangerous and existing smells that mean no harm.
“Habituation is a type of background subtraction method that computer scientists have long used in areas such as computer vision. The basic goal is to identify changes in a scene (e.g., a moving car, or a person walking), by subtracting a static or nearly static background, and therefore, highlighting the foreground. Turns out that this is an important problem in olfaction, as well,” says Navlakha.
Going forward, Navlakha tells Inverse that a simple biologically derived algorithm like this could be an important tool for computer systems, like neural networks, that process a lot of information very quickly.
“[The] approach we developed studying the fruit fly is very simple, easy to implement, and has good properties (e.g., fast dishabituation in case the background becomes important),” says Navlakha. “As an example, we showed in our paper that this type of algorithm could be useful for performing online similarity searches. We also think it may play well in deep networks, though we did not explore this idea here.”
And even beyond simple computer processing, Navlakha says that this algorithm could be used to develop electronic noses that can sniff out chemical weapons or dangerous pollution.
“There are lots of potential applications for electronic noses, ranging from pollution detection in cities, quality control of food in kitchens and restaurants, and chemical detection in combat scenarios,” says Navlakha. “This would be similar to giving robots a sense of smell, and there are indeed many companies today trying to replicate some of the basic features of the nose (ranging from designing olfactory sensors to designing odor identification algorithms).”
inverse.com, 12 May 2020