Just like a fruit fly, a new algorithm never forgets old scents

Researchers from the Okinawa Institute of Science and Technology have introduced a biological-inspired algorithm, Spi-Fly, designed to improve how artificial olfaction systems identify and remember different odors.

Traditional electronic noses often suffer from narrow functionality and catastrophic forgetting, meaning they lose previously acquired data when trained on a new scent. To overcome this limitation, the researchers modeled their software on the olfactory system of the fruit fly, which uses a process known as sparse coding to generate distinctive neural "barcodes" for complex sensory inputs. This allows the insect to recognize a wide range of smells with an incredibly small number of active neurons, bypassing the need for intensive computational training methods like backpropagation.

While Spi-Fly demonstrated superior performance in simulations compared to traditional training techniques—requiring fewer exposures to learn new smells while resisting memory loss—it still faces several practical hurdles. Critics point out that the algorithm has only been tested against simulated, isolated inputs rather than the complex, overlapping mixtures found in real-world environments. Furthermore, other researchers argue that the model may not be directly comparable to alternative systems designed for different noise-filtering tasks. The developers plan to address these limitations by transitioning the purely digital algorithm onto actual physical neuromorphic hardware to test its viability under real-world conditions.